├── .gitignore ├── LICENSE ├── README.md ├── assets └── graphrec.png ├── dataloader.py ├── datasets ├── Ciao │ ├── rating.mat │ └── trustnetwork.mat └── Epinions │ ├── ratings_data.txt │ └── trust_data.txt ├── main.py ├── model.py ├── preprocess.py ├── requirements.txt └── utils.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | *.egg-info/ 24 | .installed.cfg 25 | *.egg 26 | MANIFEST 27 | 28 | # PyInstaller 29 | # Usually these files are written by a python script from a template 30 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 31 | *.manifest 32 | *.spec 33 | 34 | # Installer logs 35 | pip-log.txt 36 | pip-delete-this-directory.txt 37 | 38 | # Unit test / coverage reports 39 | htmlcov/ 40 | .tox/ 41 | .coverage 42 | .coverage.* 43 | .cache 44 | nosetests.xml 45 | coverage.xml 46 | *.cover 47 | .hypothesis/ 48 | .pytest_cache/ 49 | 50 | # Translations 51 | *.mo 52 | *.pot 53 | 54 | # Django stuff: 55 | *.log 56 | local_settings.py 57 | db.sqlite3 58 | 59 | # Flask stuff: 60 | instance/ 61 | .webassets-cache 62 | 63 | # Scrapy stuff: 64 | .scrapy 65 | 66 | # Sphinx documentation 67 | docs/_build/ 68 | 69 | # PyBuilder 70 | target/ 71 | 72 | # Jupyter Notebook 73 | .ipynb_checkpoints 74 | 75 | # pyenv 76 | .python-version 77 | 78 | # celery beat schedule file 79 | celerybeat-schedule 80 | 81 | # SageMath parsed files 82 | *.sage.py 83 | 84 | # Environments 85 | .env 86 | .venv 87 | env/ 88 | venv/ 89 | ENV/ 90 | env.bak/ 91 | venv.bak/ 92 | 93 | # Spyder project settings 94 | .spyderproject 95 | .spyproject 96 | 97 | # Rope project settings 98 | .ropeproject 99 | 100 | # mkdocs documentation 101 | /site 102 | 103 | # mypy 104 | .mypy_cache/ 105 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | GNU GENERAL PUBLIC LICENSE 2 | Version 3, 29 June 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU General Public License is a free, copyleft license for 11 | software and other kinds of works. 12 | 13 | The licenses for most software and other practical works are designed 14 | to take away your freedom to share and change the works. By contrast, 15 | the GNU General Public License is intended to guarantee your freedom to 16 | share and change all versions of a program--to make sure it remains free 17 | software for all its users. We, the Free Software Foundation, use the 18 | GNU General Public License for most of our software; it applies also to 19 | any other work released this way by its authors. You can apply it to 20 | your programs, too. 21 | 22 | When we speak of free software, we are referring to freedom, not 23 | price. Our General Public Licenses are designed to make sure that you 24 | have the freedom to distribute copies of free software (and charge for 25 | them if you wish), that you receive source code or can get it if you 26 | want it, that you can change the software or use pieces of it in new 27 | free programs, and that you know you can do these things. 28 | 29 | To protect your rights, we need to prevent others from denying you 30 | these rights or asking you to surrender the rights. Therefore, you have 31 | certain responsibilities if you distribute copies of the software, or if 32 | you modify it: responsibilities to respect the freedom of others. 33 | 34 | For example, if you distribute copies of such a program, whether 35 | gratis or for a fee, you must pass on to the recipients the same 36 | freedoms that you received. You must make sure that they, too, receive 37 | or can get the source code. And you must show them these terms so they 38 | know their rights. 39 | 40 | Developers that use the GNU GPL protect your rights with two steps: 41 | (1) assert copyright on the software, and (2) offer you this License 42 | giving you legal permission to copy, distribute and/or modify it. 43 | 44 | For the developers' and authors' protection, the GPL clearly explains 45 | that there is no warranty for this free software. For both users' and 46 | authors' sake, the GPL requires that modified versions be marked as 47 | changed, so that their problems will not be attributed erroneously to 48 | authors of previous versions. 49 | 50 | Some devices are designed to deny users access to install or run 51 | modified versions of the software inside them, although the manufacturer 52 | can do so. This is fundamentally incompatible with the aim of 53 | protecting users' freedom to change the software. The systematic 54 | pattern of such abuse occurs in the area of products for individuals to 55 | use, which is precisely where it is most unacceptable. Therefore, we 56 | have designed this version of the GPL to prohibit the practice for those 57 | products. If such problems arise substantially in other domains, we 58 | stand ready to extend this provision to those domains in future versions 59 | of the GPL, as needed to protect the freedom of users. 60 | 61 | Finally, every program is threatened constantly by software patents. 62 | States should not allow patents to restrict development and use of 63 | software on general-purpose computers, but in those that do, we wish to 64 | avoid the special danger that patents applied to a free program could 65 | make it effectively proprietary. To prevent this, the GPL assures that 66 | patents cannot be used to render the program non-free. 67 | 68 | The precise terms and conditions for copying, distribution and 69 | modification follow. 70 | 71 | TERMS AND CONDITIONS 72 | 73 | 0. Definitions. 74 | 75 | "This License" refers to version 3 of the GNU General Public License. 76 | 77 | "Copyright" also means copyright-like laws that apply to other kinds of 78 | works, such as semiconductor masks. 79 | 80 | "The Program" refers to any copyrightable work licensed under this 81 | License. Each licensee is addressed as "you". "Licensees" and 82 | "recipients" may be individuals or organizations. 83 | 84 | To "modify" a work means to copy from or adapt all or part of the work 85 | in a fashion requiring copyright permission, other than the making of an 86 | exact copy. The resulting work is called a "modified version" of the 87 | earlier work or a work "based on" the earlier work. 88 | 89 | A "covered work" means either the unmodified Program or a work based 90 | on the Program. 91 | 92 | To "propagate" a work means to do anything with it that, without 93 | permission, would make you directly or secondarily liable for 94 | infringement under applicable copyright law, except executing it on a 95 | computer or modifying a private copy. Propagation includes copying, 96 | distribution (with or without modification), making available to the 97 | public, and in some countries other activities as well. 98 | 99 | To "convey" a work means any kind of propagation that enables other 100 | parties to make or receive copies. Mere interaction with a user through 101 | a computer network, with no transfer of a copy, is not conveying. 102 | 103 | An interactive user interface displays "Appropriate Legal Notices" 104 | to the extent that it includes a convenient and prominently visible 105 | feature that (1) displays an appropriate copyright notice, and (2) 106 | tells the user that there is no warranty for the work (except to the 107 | extent that warranties are provided), that licensees may convey the 108 | work under this License, and how to view a copy of this License. If 109 | the interface presents a list of user commands or options, such as a 110 | menu, a prominent item in the list meets this criterion. 111 | 112 | 1. Source Code. 113 | 114 | The "source code" for a work means the preferred form of the work 115 | for making modifications to it. "Object code" means any non-source 116 | form of a work. 117 | 118 | A "Standard Interface" means an interface that either is an official 119 | standard defined by a recognized standards body, or, in the case of 120 | interfaces specified for a particular programming language, one that 121 | is widely used among developers working in that language. 122 | 123 | The "System Libraries" of an executable work include anything, other 124 | than the work as a whole, that (a) is included in the normal form of 125 | packaging a Major Component, but which is not part of that Major 126 | Component, and (b) serves only to enable use of the work with that 127 | Major Component, or to implement a Standard Interface for which an 128 | implementation is available to the public in source code form. A 129 | "Major Component", in this context, means a major essential component 130 | (kernel, window system, and so on) of the specific operating system 131 | (if any) on which the executable work runs, or a compiler used to 132 | produce the work, or an object code interpreter used to run it. 133 | 134 | The "Corresponding Source" for a work in object code form means all 135 | the source code needed to generate, install, and (for an executable 136 | work) run the object code and to modify the work, including scripts to 137 | control those activities. However, it does not include the work's 138 | System Libraries, or general-purpose tools or generally available free 139 | programs which are used unmodified in performing those activities but 140 | which are not part of the work. For example, Corresponding Source 141 | includes interface definition files associated with source files for 142 | the work, and the source code for shared libraries and dynamically 143 | linked subprograms that the work is specifically designed to require, 144 | such as by intimate data communication or control flow between those 145 | subprograms and other parts of the work. 146 | 147 | The Corresponding Source need not include anything that users 148 | can regenerate automatically from other parts of the Corresponding 149 | Source. 150 | 151 | The Corresponding Source for a work in source code form is that 152 | same work. 153 | 154 | 2. Basic Permissions. 155 | 156 | All rights granted under this License are granted for the term of 157 | copyright on the Program, and are irrevocable provided the stated 158 | conditions are met. This License explicitly affirms your unlimited 159 | permission to run the unmodified Program. The output from running a 160 | covered work is covered by this License only if the output, given its 161 | content, constitutes a covered work. This License acknowledges your 162 | rights of fair use or other equivalent, as provided by copyright law. 163 | 164 | You may make, run and propagate covered works that you do not 165 | convey, without conditions so long as your license otherwise remains 166 | in force. You may convey covered works to others for the sole purpose 167 | of having them make modifications exclusively for you, or provide you 168 | with facilities for running those works, provided that you comply with 169 | the terms of this License in conveying all material for which you do 170 | not control copyright. Those thus making or running the covered works 171 | for you must do so exclusively on your behalf, under your direction 172 | and control, on terms that prohibit them from making any copies of 173 | your copyrighted material outside their relationship with you. 174 | 175 | Conveying under any other circumstances is permitted solely under 176 | the conditions stated below. Sublicensing is not allowed; section 10 177 | makes it unnecessary. 178 | 179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 180 | 181 | No covered work shall be deemed part of an effective technological 182 | measure under any applicable law fulfilling obligations under article 183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 184 | similar laws prohibiting or restricting circumvention of such 185 | measures. 186 | 187 | When you convey a covered work, you waive any legal power to forbid 188 | circumvention of technological measures to the extent such circumvention 189 | is effected by exercising rights under this License with respect to 190 | the covered work, and you disclaim any intention to limit operation or 191 | modification of the work as a means of enforcing, against the work's 192 | users, your or third parties' legal rights to forbid circumvention of 193 | technological measures. 194 | 195 | 4. Conveying Verbatim Copies. 196 | 197 | You may convey verbatim copies of the Program's source code as you 198 | receive it, in any medium, provided that you conspicuously and 199 | appropriately publish on each copy an appropriate copyright notice; 200 | keep intact all notices stating that this License and any 201 | non-permissive terms added in accord with section 7 apply to the code; 202 | keep intact all notices of the absence of any warranty; and give all 203 | recipients a copy of this License along with the Program. 204 | 205 | You may charge any price or no price for each copy that you convey, 206 | and you may offer support or warranty protection for a fee. 207 | 208 | 5. Conveying Modified Source Versions. 209 | 210 | You may convey a work based on the Program, or the modifications to 211 | produce it from the Program, in the form of source code under the 212 | terms of section 4, provided that you also meet all of these conditions: 213 | 214 | a) The work must carry prominent notices stating that you modified 215 | it, and giving a relevant date. 216 | 217 | b) The work must carry prominent notices stating that it is 218 | released under this License and any conditions added under section 219 | 7. This requirement modifies the requirement in section 4 to 220 | "keep intact all notices". 221 | 222 | c) You must license the entire work, as a whole, under this 223 | License to anyone who comes into possession of a copy. This 224 | License will therefore apply, along with any applicable section 7 225 | additional terms, to the whole of the work, and all its parts, 226 | regardless of how they are packaged. This License gives no 227 | permission to license the work in any other way, but it does not 228 | invalidate such permission if you have separately received it. 229 | 230 | d) If the work has interactive user interfaces, each must display 231 | Appropriate Legal Notices; however, if the Program has interactive 232 | interfaces that do not display Appropriate Legal Notices, your 233 | work need not make them do so. 234 | 235 | A compilation of a covered work with other separate and independent 236 | works, which are not by their nature extensions of the covered work, 237 | and which are not combined with it such as to form a larger program, 238 | in or on a volume of a storage or distribution medium, is called an 239 | "aggregate" if the compilation and its resulting copyright are not 240 | used to limit the access or legal rights of the compilation's users 241 | beyond what the individual works permit. Inclusion of a covered work 242 | in an aggregate does not cause this License to apply to the other 243 | parts of the aggregate. 244 | 245 | 6. Conveying Non-Source Forms. 246 | 247 | You may convey a covered work in object code form under the terms 248 | of sections 4 and 5, provided that you also convey the 249 | machine-readable Corresponding Source under the terms of this License, 250 | in one of these ways: 251 | 252 | a) Convey the object code in, or embodied in, a physical product 253 | (including a physical distribution medium), accompanied by the 254 | Corresponding Source fixed on a durable physical medium 255 | customarily used for software interchange. 256 | 257 | b) Convey the object code in, or embodied in, a physical product 258 | (including a physical distribution medium), accompanied by a 259 | written offer, valid for at least three years and valid for as 260 | long as you offer spare parts or customer support for that product 261 | model, to give anyone who possesses the object code either (1) a 262 | copy of the Corresponding Source for all the software in the 263 | product that is covered by this License, on a durable physical 264 | medium customarily used for software interchange, for a price no 265 | more than your reasonable cost of physically performing this 266 | conveying of source, or (2) access to copy the 267 | Corresponding Source from a network server at no charge. 268 | 269 | c) Convey individual copies of the object code with a copy of the 270 | written offer to provide the Corresponding Source. This 271 | alternative is allowed only occasionally and noncommercially, and 272 | only if you received the object code with such an offer, in accord 273 | with subsection 6b. 274 | 275 | d) Convey the object code by offering access from a designated 276 | place (gratis or for a charge), and offer equivalent access to the 277 | Corresponding Source in the same way through the same place at no 278 | further charge. You need not require recipients to copy the 279 | Corresponding Source along with the object code. If the place to 280 | copy the object code is a network server, the Corresponding Source 281 | may be on a different server (operated by you or a third party) 282 | that supports equivalent copying facilities, provided you maintain 283 | clear directions next to the object code saying where to find the 284 | Corresponding Source. Regardless of what server hosts the 285 | Corresponding Source, you remain obligated to ensure that it is 286 | available for as long as needed to satisfy these requirements. 287 | 288 | e) Convey the object code using peer-to-peer transmission, provided 289 | you inform other peers where the object code and Corresponding 290 | Source of the work are being offered to the general public at no 291 | charge under subsection 6d. 292 | 293 | A separable portion of the object code, whose source code is excluded 294 | from the Corresponding Source as a System Library, need not be 295 | included in conveying the object code work. 296 | 297 | A "User Product" is either (1) a "consumer product", which means any 298 | tangible personal property which is normally used for personal, family, 299 | or household purposes, or (2) anything designed or sold for incorporation 300 | into a dwelling. In determining whether a product is a consumer product, 301 | doubtful cases shall be resolved in favor of coverage. For a particular 302 | product received by a particular user, "normally used" refers to a 303 | typical or common use of that class of product, regardless of the status 304 | of the particular user or of the way in which the particular user 305 | actually uses, or expects or is expected to use, the product. A product 306 | is a consumer product regardless of whether the product has substantial 307 | commercial, industrial or non-consumer uses, unless such uses represent 308 | the only significant mode of use of the product. 309 | 310 | "Installation Information" for a User Product means any methods, 311 | procedures, authorization keys, or other information required to install 312 | and execute modified versions of a covered work in that User Product from 313 | a modified version of its Corresponding Source. The information must 314 | suffice to ensure that the continued functioning of the modified object 315 | code is in no case prevented or interfered with solely because 316 | modification has been made. 317 | 318 | If you convey an object code work under this section in, or with, or 319 | specifically for use in, a User Product, and the conveying occurs as 320 | part of a transaction in which the right of possession and use of the 321 | User Product is transferred to the recipient in perpetuity or for a 322 | fixed term (regardless of how the transaction is characterized), the 323 | Corresponding Source conveyed under this section must be accompanied 324 | by the Installation Information. But this requirement does not apply 325 | if neither you nor any third party retains the ability to install 326 | modified object code on the User Product (for example, the work has 327 | been installed in ROM). 328 | 329 | The requirement to provide Installation Information does not include a 330 | requirement to continue to provide support service, warranty, or updates 331 | for a work that has been modified or installed by the recipient, or for 332 | the User Product in which it has been modified or installed. Access to a 333 | network may be denied when the modification itself materially and 334 | adversely affects the operation of the network or violates the rules and 335 | protocols for communication across the network. 336 | 337 | Corresponding Source conveyed, and Installation Information provided, 338 | in accord with this section must be in a format that is publicly 339 | documented (and with an implementation available to the public in 340 | source code form), and must require no special password or key for 341 | unpacking, reading or copying. 342 | 343 | 7. Additional Terms. 344 | 345 | "Additional permissions" are terms that supplement the terms of this 346 | License by making exceptions from one or more of its conditions. 347 | Additional permissions that are applicable to the entire Program shall 348 | be treated as though they were included in this License, to the extent 349 | that they are valid under applicable law. If additional permissions 350 | apply only to part of the Program, that part may be used separately 351 | under those permissions, but the entire Program remains governed by 352 | this License without regard to the additional permissions. 353 | 354 | When you convey a copy of a covered work, you may at your option 355 | remove any additional permissions from that copy, or from any part of 356 | it. (Additional permissions may be written to require their own 357 | removal in certain cases when you modify the work.) You may place 358 | additional permissions on material, added by you to a covered work, 359 | for which you have or can give appropriate copyright permission. 360 | 361 | Notwithstanding any other provision of this License, for material you 362 | add to a covered work, you may (if authorized by the copyright holders of 363 | that material) supplement the terms of this License with terms: 364 | 365 | a) Disclaiming warranty or limiting liability differently from the 366 | terms of sections 15 and 16 of this License; or 367 | 368 | b) Requiring preservation of specified reasonable legal notices or 369 | author attributions in that material or in the Appropriate Legal 370 | Notices displayed by works containing it; or 371 | 372 | c) Prohibiting misrepresentation of the origin of that material, or 373 | requiring that modified versions of such material be marked in 374 | reasonable ways as different from the original version; or 375 | 376 | d) Limiting the use for publicity purposes of names of licensors or 377 | authors of the material; or 378 | 379 | e) Declining to grant rights under trademark law for use of some 380 | trade names, trademarks, or service marks; or 381 | 382 | f) Requiring indemnification of licensors and authors of that 383 | material by anyone who conveys the material (or modified versions of 384 | it) with contractual assumptions of liability to the recipient, for 385 | any liability that these contractual assumptions directly impose on 386 | those licensors and authors. 387 | 388 | All other non-permissive additional terms are considered "further 389 | restrictions" within the meaning of section 10. If the Program as you 390 | received it, or any part of it, contains a notice stating that it is 391 | governed by this License along with a term that is a further 392 | restriction, you may remove that term. If a license document contains 393 | a further restriction but permits relicensing or conveying under this 394 | License, you may add to a covered work material governed by the terms 395 | of that license document, provided that the further restriction does 396 | not survive such relicensing or conveying. 397 | 398 | If you add terms to a covered work in accord with this section, you 399 | must place, in the relevant source files, a statement of the 400 | additional terms that apply to those files, or a notice indicating 401 | where to find the applicable terms. 402 | 403 | Additional terms, permissive or non-permissive, may be stated in the 404 | form of a separately written license, or stated as exceptions; 405 | the above requirements apply either way. 406 | 407 | 8. Termination. 408 | 409 | You may not propagate or modify a covered work except as expressly 410 | provided under this License. Any attempt otherwise to propagate or 411 | modify it is void, and will automatically terminate your rights under 412 | this License (including any patent licenses granted under the third 413 | paragraph of section 11). 414 | 415 | However, if you cease all violation of this License, then your 416 | license from a particular copyright holder is reinstated (a) 417 | provisionally, unless and until the copyright holder explicitly and 418 | finally terminates your license, and (b) permanently, if the copyright 419 | holder fails to notify you of the violation by some reasonable means 420 | prior to 60 days after the cessation. 421 | 422 | Moreover, your license from a particular copyright holder is 423 | reinstated permanently if the copyright holder notifies you of the 424 | violation by some reasonable means, this is the first time you have 425 | received notice of violation of this License (for any work) from that 426 | copyright holder, and you cure the violation prior to 30 days after 427 | your receipt of the notice. 428 | 429 | Termination of your rights under this section does not terminate the 430 | licenses of parties who have received copies or rights from you under 431 | this License. If your rights have been terminated and not permanently 432 | reinstated, you do not qualify to receive new licenses for the same 433 | material under section 10. 434 | 435 | 9. Acceptance Not Required for Having Copies. 436 | 437 | You are not required to accept this License in order to receive or 438 | run a copy of the Program. Ancillary propagation of a covered work 439 | occurring solely as a consequence of using peer-to-peer transmission 440 | to receive a copy likewise does not require acceptance. However, 441 | nothing other than this License grants you permission to propagate or 442 | modify any covered work. These actions infringe copyright if you do 443 | not accept this License. Therefore, by modifying or propagating a 444 | covered work, you indicate your acceptance of this License to do so. 445 | 446 | 10. Automatic Licensing of Downstream Recipients. 447 | 448 | Each time you convey a covered work, the recipient automatically 449 | receives a license from the original licensors, to run, modify and 450 | propagate that work, subject to this License. You are not responsible 451 | for enforcing compliance by third parties with this License. 452 | 453 | An "entity transaction" is a transaction transferring control of an 454 | organization, or substantially all assets of one, or subdividing an 455 | organization, or merging organizations. If propagation of a covered 456 | work results from an entity transaction, each party to that 457 | transaction who receives a copy of the work also receives whatever 458 | licenses to the work the party's predecessor in interest had or could 459 | give under the previous paragraph, plus a right to possession of the 460 | Corresponding Source of the work from the predecessor in interest, if 461 | the predecessor has it or can get it with reasonable efforts. 462 | 463 | You may not impose any further restrictions on the exercise of the 464 | rights granted or affirmed under this License. For example, you may 465 | not impose a license fee, royalty, or other charge for exercise of 466 | rights granted under this License, and you may not initiate litigation 467 | (including a cross-claim or counterclaim in a lawsuit) alleging that 468 | any patent claim is infringed by making, using, selling, offering for 469 | sale, or importing the Program or any portion of it. 470 | 471 | 11. Patents. 472 | 473 | A "contributor" is a copyright holder who authorizes use under this 474 | License of the Program or a work on which the Program is based. The 475 | work thus licensed is called the contributor's "contributor version". 476 | 477 | A contributor's "essential patent claims" are all patent claims 478 | owned or controlled by the contributor, whether already acquired or 479 | hereafter acquired, that would be infringed by some manner, permitted 480 | by this License, of making, using, or selling its contributor version, 481 | but do not include claims that would be infringed only as a 482 | consequence of further modification of the contributor version. For 483 | purposes of this definition, "control" includes the right to grant 484 | patent sublicenses in a manner consistent with the requirements of 485 | this License. 486 | 487 | Each contributor grants you a non-exclusive, worldwide, royalty-free 488 | patent license under the contributor's essential patent claims, to 489 | make, use, sell, offer for sale, import and otherwise run, modify and 490 | propagate the contents of its contributor version. 491 | 492 | In the following three paragraphs, a "patent license" is any express 493 | agreement or commitment, however denominated, not to enforce a patent 494 | (such as an express permission to practice a patent or covenant not to 495 | sue for patent infringement). To "grant" such a patent license to a 496 | party means to make such an agreement or commitment not to enforce a 497 | patent against the party. 498 | 499 | If you convey a covered work, knowingly relying on a patent license, 500 | and the Corresponding Source of the work is not available for anyone 501 | to copy, free of charge and under the terms of this License, through a 502 | publicly available network server or other readily accessible means, 503 | then you must either (1) cause the Corresponding Source to be so 504 | available, or (2) arrange to deprive yourself of the benefit of the 505 | patent license for this particular work, or (3) arrange, in a manner 506 | consistent with the requirements of this License, to extend the patent 507 | license to downstream recipients. "Knowingly relying" means you have 508 | actual knowledge that, but for the patent license, your conveying the 509 | covered work in a country, or your recipient's use of the covered work 510 | in a country, would infringe one or more identifiable patents in that 511 | country that you have reason to believe are valid. 512 | 513 | If, pursuant to or in connection with a single transaction or 514 | arrangement, you convey, or propagate by procuring conveyance of, a 515 | covered work, and grant a patent license to some of the parties 516 | receiving the covered work authorizing them to use, propagate, modify 517 | or convey a specific copy of the covered work, then the patent license 518 | you grant is automatically extended to all recipients of the covered 519 | work and works based on it. 520 | 521 | A patent license is "discriminatory" if it does not include within 522 | the scope of its coverage, prohibits the exercise of, or is 523 | conditioned on the non-exercise of one or more of the rights that are 524 | specifically granted under this License. You may not convey a covered 525 | work if you are a party to an arrangement with a third party that is 526 | in the business of distributing software, under which you make payment 527 | to the third party based on the extent of your activity of conveying 528 | the work, and under which the third party grants, to any of the 529 | parties who would receive the covered work from you, a discriminatory 530 | patent license (a) in connection with copies of the covered work 531 | conveyed by you (or copies made from those copies), or (b) primarily 532 | for and in connection with specific products or compilations that 533 | contain the covered work, unless you entered into that arrangement, 534 | or that patent license was granted, prior to 28 March 2007. 535 | 536 | Nothing in this License shall be construed as excluding or limiting 537 | any implied license or other defenses to infringement that may 538 | otherwise be available to you under applicable patent law. 539 | 540 | 12. No Surrender of Others' Freedom. 541 | 542 | If conditions are imposed on you (whether by court order, agreement or 543 | otherwise) that contradict the conditions of this License, they do not 544 | excuse you from the conditions of this License. If you cannot convey a 545 | covered work so as to satisfy simultaneously your obligations under this 546 | License and any other pertinent obligations, then as a consequence you may 547 | not convey it at all. For example, if you agree to terms that obligate you 548 | to collect a royalty for further conveying from those to whom you convey 549 | the Program, the only way you could satisfy both those terms and this 550 | License would be to refrain entirely from conveying the Program. 551 | 552 | 13. Use with the GNU Affero General Public License. 553 | 554 | Notwithstanding any other provision of this License, you have 555 | permission to link or combine any covered work with a work licensed 556 | under version 3 of the GNU Affero General Public License into a single 557 | combined work, and to convey the resulting work. The terms of this 558 | License will continue to apply to the part which is the covered work, 559 | but the special requirements of the GNU Affero General Public License, 560 | section 13, concerning interaction through a network will apply to the 561 | combination as such. 562 | 563 | 14. Revised Versions of this License. 564 | 565 | The Free Software Foundation may publish revised and/or new versions of 566 | the GNU General Public License from time to time. Such new versions will 567 | be similar in spirit to the present version, but may differ in detail to 568 | address new problems or concerns. 569 | 570 | Each version is given a distinguishing version number. If the 571 | Program specifies that a certain numbered version of the GNU General 572 | Public License "or any later version" applies to it, you have the 573 | option of following the terms and conditions either of that numbered 574 | version or of any later version published by the Free Software 575 | Foundation. If the Program does not specify a version number of the 576 | GNU General Public License, you may choose any version ever published 577 | by the Free Software Foundation. 578 | 579 | If the Program specifies that a proxy can decide which future 580 | versions of the GNU General Public License can be used, that proxy's 581 | public statement of acceptance of a version permanently authorizes you 582 | to choose that version for the Program. 583 | 584 | Later license versions may give you additional or different 585 | permissions. However, no additional obligations are imposed on any 586 | author or copyright holder as a result of your choosing to follow a 587 | later version. 588 | 589 | 15. Disclaimer of Warranty. 590 | 591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 599 | 600 | 16. Limitation of Liability. 601 | 602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 610 | SUCH DAMAGES. 611 | 612 | 17. Interpretation of Sections 15 and 16. 613 | 614 | If the disclaimer of warranty and limitation of liability provided 615 | above cannot be given local legal effect according to their terms, 616 | reviewing courts shall apply local law that most closely approximates 617 | an absolute waiver of all civil liability in connection with the 618 | Program, unless a warranty or assumption of liability accompanies a 619 | copy of the Program in return for a fee. 620 | 621 | END OF TERMS AND CONDITIONS 622 | 623 | How to Apply These Terms to Your New Programs 624 | 625 | If you develop a new program, and you want it to be of the greatest 626 | possible use to the public, the best way to achieve this is to make it 627 | free software which everyone can redistribute and change under these terms. 628 | 629 | To do so, attach the following notices to the program. It is safest 630 | to attach them to the start of each source file to most effectively 631 | state the exclusion of warranty; and each file should have at least 632 | the "copyright" line and a pointer to where the full notice is found. 633 | 634 | 635 | Copyright (C) 636 | 637 | This program is free software: you can redistribute it and/or modify 638 | it under the terms of the GNU General Public License as published by 639 | the Free Software Foundation, either version 3 of the License, or 640 | (at your option) any later version. 641 | 642 | This program is distributed in the hope that it will be useful, 643 | but WITHOUT ANY WARRANTY; without even the implied warranty of 644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 645 | GNU General Public License for more details. 646 | 647 | You should have received a copy of the GNU General Public License 648 | along with this program. If not, see . 649 | 650 | Also add information on how to contact you by electronic and paper mail. 651 | 652 | If the program does terminal interaction, make it output a short 653 | notice like this when it starts in an interactive mode: 654 | 655 | Copyright (C) 656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. 657 | This is free software, and you are welcome to redistribute it 658 | under certain conditions; type `show c' for details. 659 | 660 | The hypothetical commands `show w' and `show c' should show the appropriate 661 | parts of the General Public License. Of course, your program's commands 662 | might be different; for a GUI interface, you would use an "about box". 663 | 664 | You should also get your employer (if you work as a programmer) or school, 665 | if any, to sign a "copyright disclaimer" for the program, if necessary. 666 | For more information on this, and how to apply and follow the GNU GPL, see 667 | . 668 | 669 | The GNU General Public License does not permit incorporating your program 670 | into proprietary programs. If your program is a subroutine library, you 671 | may consider it more useful to permit linking proprietary applications with 672 | the library. If this is what you want to do, use the GNU Lesser General 673 | Public License instead of this License. But first, please read 674 | . 675 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # GraphRec_PyTorch 2 | A PyTorch implementation of the GraphRec model in [Graph Neural Networks for Social Recommendation](https://arxiv.org/pdf/1902.07243.pdf) (Fan, Wenqi, et al. "Graph Neural Networks for Social Recommendation." The World Wide Web Conference. ACM, 2019). 3 | 4 | ![architecture](assets/graphrec.png) 5 | 6 | 7 | # Usage 8 | 9 | 1. Install required packages from requirements.txt file. 10 | ```bash 11 | pip install -r requirements.txt 12 | ``` 13 | 14 | 2. Preprocess dataset. Two pkl files named dataset and list should be generated in the respective folders of the dataset. 15 | ```bash 16 | python preprocess.py --dataset Ciao 17 | python preprocess.py --dataset Epinions 18 | ``` 19 | 20 | 3. Run main.py file to train the model. You can configure some training parameters through the command line. 21 | ```bash 22 | python main.py 23 | ``` 24 | 25 | 4. Run main.py file to test the model. 26 | ```bash 27 | python main.py --test 28 | ``` 29 | 30 | -------------------------------------------------------------------------------- /assets/graphrec.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wang-Shuo/GraphRec_PyTorch/5768c61f66b238dc057c09535b00eec687daac2c/assets/graphrec.png -------------------------------------------------------------------------------- /dataloader.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import random 3 | import torch 4 | from torch.utils.data import Dataset 5 | 6 | class GRDataset(Dataset): 7 | def __init__(self, data, u_items_list, u_users_list, u_users_items_list, i_users_list): 8 | self.data = data 9 | self.u_items_list = u_items_list 10 | self.u_users_list = u_users_list 11 | self.u_users_items_list = u_users_items_list 12 | self.i_users_list = i_users_list 13 | 14 | def __getitem__(self, index): 15 | uid = self.data[index][0] 16 | iid = self.data[index][1] 17 | label = self.data[index][2] 18 | u_items = self.u_items_list[uid] 19 | u_users = self.u_users_list[uid] 20 | u_users_items = self.u_users_items_list[uid] 21 | i_users = self.i_users_list[iid] 22 | 23 | return (uid, iid, label), u_items, u_users, u_users_items, i_users 24 | 25 | def __len__(self): 26 | return len(self.data) 27 | -------------------------------------------------------------------------------- /datasets/Ciao/rating.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wang-Shuo/GraphRec_PyTorch/5768c61f66b238dc057c09535b00eec687daac2c/datasets/Ciao/rating.mat -------------------------------------------------------------------------------- /datasets/Ciao/trustnetwork.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wang-Shuo/GraphRec_PyTorch/5768c61f66b238dc057c09535b00eec687daac2c/datasets/Ciao/trustnetwork.mat -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python37 2 | # -*- coding: utf-8 -*- 3 | """ 4 | Created on 30 Sep, 2019 5 | 6 | @author: wangshuo 7 | """ 8 | 9 | import os 10 | import time 11 | import argparse 12 | import pickle 13 | import numpy as np 14 | import random 15 | from tqdm import tqdm 16 | from os.path import join 17 | 18 | import torch 19 | from torch import nn 20 | from torch.utils.data import DataLoader 21 | import torch.nn.functional as F 22 | import torch.optim as optim 23 | from torch.optim.lr_scheduler import StepLR 24 | from torch.autograd import Variable 25 | from torch.backends import cudnn 26 | 27 | from utils import collate_fn 28 | from model import GraphRec 29 | from dataloader import GRDataset 30 | 31 | parser = argparse.ArgumentParser() 32 | parser.add_argument('--dataset_path', default='dataset/Ciao/', help='dataset directory path: datasets/Ciao/Epinions') 33 | parser.add_argument('--batch_size', type=int, default=256, help='input batch size') 34 | parser.add_argument('--embed_dim', type=int, default=64, help='the dimension of embedding') 35 | parser.add_argument('--epoch', type=int, default=30, help='the number of epochs to train for') 36 | parser.add_argument('--lr', type=float, default=0.001, help='learning rate') # [0.001, 0.0005, 0.0001] 37 | parser.add_argument('--lr_dc', type=float, default=0.1, help='learning rate decay rate') 38 | parser.add_argument('--lr_dc_step', type=int, default=30, help='the number of steps after which the learning rate decay') 39 | parser.add_argument('--test', action='store_true', help='test') 40 | args = parser.parse_args() 41 | print(args) 42 | 43 | here = os.path.dirname(os.path.abspath(__file__)) 44 | device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') 45 | 46 | def main(): 47 | print('Loading data...') 48 | with open(args.dataset_path + 'dataset.pkl', 'rb') as f: 49 | train_set = pickle.load(f) 50 | valid_set = pickle.load(f) 51 | test_set = pickle.load(f) 52 | 53 | with open(args.dataset_path + 'list.pkl', 'rb') as f: 54 | u_items_list = pickle.load(f) 55 | u_users_list = pickle.load(f) 56 | u_users_items_list = pickle.load(f) 57 | i_users_list = pickle.load(f) 58 | (user_count, item_count, rate_count) = pickle.load(f) 59 | 60 | train_data = GRDataset(train_set, u_items_list, u_users_list, u_users_items_list, i_users_list) 61 | valid_data = GRDataset(valid_set, u_items_list, u_users_list, u_users_items_list, i_users_list) 62 | test_data = GRDataset(test_set, u_items_list, u_users_list, u_users_items_list, i_users_list) 63 | train_loader = DataLoader(train_data, batch_size = args.batch_size, shuffle = True, collate_fn = collate_fn) 64 | valid_loader = DataLoader(valid_data, batch_size = args.batch_size, shuffle = False, collate_fn = collate_fn) 65 | test_loader = DataLoader(test_data, batch_size = args.batch_size, shuffle = False, collate_fn = collate_fn) 66 | 67 | model = GraphRec(user_count+1, item_count+1, rate_count+1, args.embed_dim).to(device) 68 | 69 | if args.test: 70 | print('Load checkpoint and testing...') 71 | ckpt = torch.load('best_checkpoint.pth.tar') 72 | model.load_state_dict(ckpt['state_dict']) 73 | mae, rmse = validate(test_loader, model) 74 | print("Test: MAE: {:.4f}, RMSE: {:.4f}".format(mae, rmse)) 75 | return 76 | 77 | optimizer = optim.RMSprop(model.parameters(), args.lr) 78 | criterion = nn.MSELoss() 79 | scheduler = StepLR(optimizer, step_size = args.lr_dc_step, gamma = args.lr_dc) 80 | 81 | for epoch in tqdm(range(args.epoch)): 82 | # train for one epoch 83 | scheduler.step(epoch = epoch) 84 | trainForEpoch(train_loader, model, optimizer, epoch, args.epoch, criterion, log_aggr = 100) 85 | 86 | mae, rmse = validate(valid_loader, model) 87 | 88 | # store best loss and save a model checkpoint 89 | ckpt_dict = { 90 | 'epoch': epoch + 1, 91 | 'state_dict': model.state_dict(), 92 | 'optimizer': optimizer.state_dict() 93 | } 94 | 95 | torch.save(ckpt_dict, 'latest_checkpoint.pth.tar') 96 | 97 | if epoch == 0: 98 | best_mae = mae 99 | elif mae < best_mae: 100 | best_mae = mae 101 | torch.save(ckpt_dict, 'best_checkpoint.pth.tar') 102 | 103 | print('Epoch {} validation: MAE: {:.4f}, RMSE: {:.4f}, Best MAE: {:.4f}'.format(epoch, mae, rmse, best_mae)) 104 | 105 | 106 | def trainForEpoch(train_loader, model, optimizer, epoch, num_epochs, criterion, log_aggr=1): 107 | model.train() 108 | 109 | sum_epoch_loss = 0 110 | 111 | start = time.time() 112 | for i, (uids, iids, labels, u_items, u_users, u_users_items, i_users) in tqdm(enumerate(train_loader), total=len(train_loader)): 113 | uids = uids.to(device) 114 | iids = iids.to(device) 115 | labels = labels.to(device) 116 | u_items = u_items.to(device) 117 | u_users = u_users.to(device) 118 | u_users_items = u_users_items.to(device) 119 | i_users = i_users.to(device) 120 | 121 | optimizer.zero_grad() 122 | outputs = model(uids, iids, u_items, u_users, u_users_items, i_users) 123 | 124 | loss = criterion(outputs, labels.unsqueeze(1)) 125 | loss.backward() 126 | optimizer.step() 127 | 128 | loss_val = loss.item() 129 | sum_epoch_loss += loss_val 130 | 131 | iter_num = epoch * len(train_loader) + i + 1 132 | 133 | if i % log_aggr == 0: 134 | print('[TRAIN] epoch %d/%d batch loss: %.4f (avg %.4f) (%.2f im/s)' 135 | % (epoch + 1, num_epochs, loss_val, sum_epoch_loss / (i + 1), 136 | len(uids) / (time.time() - start))) 137 | 138 | start = time.time() 139 | 140 | 141 | def validate(valid_loader, model): 142 | model.eval() 143 | errors = [] 144 | with torch.no_grad(): 145 | for uids, iids, labels, u_items, u_users, u_users_items, i_users in tqdm(valid_loader): 146 | uids = uids.to(device) 147 | iids = iids.to(device) 148 | labels = labels.to(device) 149 | u_items = u_items.to(device) 150 | u_users = u_users.to(device) 151 | u_users_items = u_users_items.to(device) 152 | i_users = i_users.to(device) 153 | preds = model(uids, iids, u_items, u_users, u_users_items, i_users) 154 | error = torch.abs(preds.squeeze(1) - labels) 155 | errors.extend(error.data.cpu().numpy().tolist()) 156 | 157 | mae = np.mean(errors) 158 | rmse = np.sqrt(np.mean(np.power(errors, 2))) 159 | return mae, rmse 160 | 161 | 162 | if __name__ == '__main__': 163 | main() 164 | -------------------------------------------------------------------------------- /model.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | import torch 3 | 4 | class _MultiLayerPercep(nn.Module): 5 | def __init__(self, input_dim, output_dim): 6 | super(_MultiLayerPercep, self).__init__() 7 | self.mlp = nn.Sequential( 8 | nn.Linear(input_dim, input_dim // 2, bias=True), 9 | nn.ReLU(), 10 | nn.Linear(input_dim // 2, output_dim, bias=True), 11 | ) 12 | 13 | def forward(self, x): 14 | return self.mlp(x) 15 | 16 | 17 | class _Aggregation(nn.Module): 18 | def __init__(self, input_dim, output_dim): 19 | super(_Aggregation, self).__init__() 20 | self.aggre = nn.Sequential( 21 | nn.Linear(input_dim, output_dim, bias=True), 22 | nn.ReLU(), 23 | ) 24 | 25 | def forward(self, x): 26 | return self.aggre(x) 27 | 28 | 29 | class _UserModel(nn.Module): 30 | ''' User modeling to learn user latent factors. 31 | User modeling leverages two types aggregation: item aggregation and social aggregation 32 | ''' 33 | def __init__(self, emb_dim, user_emb, item_emb, rate_emb): 34 | super(_UserModel, self).__init__() 35 | self.user_emb = user_emb 36 | self.item_emb = item_emb 37 | self.rate_emb = rate_emb 38 | self.emb_dim = emb_dim 39 | 40 | self.g_v = _MultiLayerPercep(2 * self.emb_dim, self.emb_dim) 41 | 42 | self.user_items_att = _MultiLayerPercep(2 * self.emb_dim, 1) 43 | self.aggre_items = _Aggregation(self.emb_dim, self.emb_dim) 44 | 45 | self.user_users_att = _MultiLayerPercep(2 * self.emb_dim, 1) 46 | self.aggre_neigbors = _Aggregation(self.emb_dim, self.emb_dim) 47 | 48 | self.combine_mlp = nn.Sequential( 49 | nn.Linear(2 * self.emb_dim, self.emb_dim, bias = True), 50 | nn.ReLU(), 51 | nn.Linear(self.emb_dim, self.emb_dim, bias = True), 52 | nn.ReLU(), 53 | nn.Linear(self.emb_dim, self.emb_dim, bias = True), 54 | nn.ReLU(), 55 | ) 56 | 57 | self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') 58 | # used for preventing zero div error when calculating softmax score 59 | self.eps = 1e-10 60 | 61 | def forward(self, uids, u_item_pad, u_user_pad, u_user_item_pad): 62 | # item aggregation 63 | q_a = self.item_emb(u_item_pad[:,:,0]) # B x maxi_len x emb_dim 64 | mask_u = torch.where(u_item_pad[:,:,0] > 0, torch.tensor([1.], device=self.device), torch.tensor([0.], device=self.device)) # B x maxi_len 65 | u_item_er = self.rate_emb(u_item_pad[:,:,1]) # B x maxi_len x emb_dim 66 | 67 | x_ia = self.g_v(torch.cat([q_a, u_item_er], dim = 2).view(-1, 2 * self.emb_dim)).view(q_a.size()) # B x maxi_len x emb_dim 68 | 69 | ## calculate attention scores in item aggregation 70 | p_i = mask_u.unsqueeze(2).expand_as(x_ia) * self.user_emb(uids).unsqueeze(1).expand_as(x_ia) # B x maxi_len x emb_dim 71 | 72 | alpha = self.user_items_att(torch.cat([x_ia, p_i], dim = 2).view(-1, 2 * self.emb_dim)).view(mask_u.size()) # B x maxi_len 73 | alpha = torch.exp(alpha) * mask_u 74 | alpha = alpha / (torch.sum(alpha, 1).unsqueeze(1).expand_as(alpha) + self.eps) 75 | 76 | h_iI = self.aggre_items(torch.sum(alpha.unsqueeze(2).expand_as(x_ia) * x_ia, 1)) # B x emb_dim 77 | 78 | # social aggregation 79 | q_a_s = self.item_emb(u_user_item_pad[:,:,:,0]) # B x maxu_len x maxi_len x emb_dim 80 | mask_s = torch.where(u_user_item_pad[:,:,:,0] > 0, torch.tensor([1.], device=self.device), torch.tensor([0.], device=self.device)) # B x maxu_len x maxi_len 81 | u_user_item_er = self.rate_emb(u_user_item_pad[:,:,:,1]) # B x maxu_len x maxi_len x emb_dim 82 | 83 | x_ia_s = self.g_v(torch.cat([q_a_s, u_user_item_er], dim = 3).view(-1, 2 * self.emb_dim)).view(q_a_s.size()) # B x maxu_len x maxi_len x emb_dim 84 | 85 | p_i_s = mask_s.unsqueeze(3).expand_as(x_ia_s) * self.user_emb(u_user_pad).unsqueeze(2).expand_as(x_ia_s) # B x maxu_len x maxi_len x emb_dim 86 | 87 | alpha_s = self.user_items_att(torch.cat([x_ia_s, p_i_s], dim = 3).view(-1, 2 * self.emb_dim)).view(mask_s.size()) # B x maxu_len x maxi_len 88 | alpha_s = torch.exp(alpha_s) * mask_s 89 | alpha_s = alpha_s / (torch.sum(alpha_s, 2).unsqueeze(2).expand_as(alpha_s) + self.eps) 90 | 91 | h_oI_temp = torch.sum(alpha_s.unsqueeze(3).expand_as(x_ia_s) * x_ia_s, 2) # B x maxu_len x emb_dim 92 | h_oI = self.aggre_items(h_oI_temp.view(-1, self.emb_dim)).view(h_oI_temp.size()) # B x maxu_len x emb_dim 93 | 94 | ## calculate attention scores in social aggregation 95 | beta = self.user_users_att(torch.cat([h_oI, self.user_emb(u_user_pad)], dim = 2).view(-1, 2 * self.emb_dim)).view(u_user_pad.size()) 96 | mask_su = torch.where(u_user_pad > 0, torch.tensor([1.], device=self.device), torch.tensor([0.], device=self.device)) 97 | beta = torch.exp(beta) * mask_su 98 | beta = beta / (torch.sum(beta, 1).unsqueeze(1).expand_as(beta) + self.eps) 99 | h_iS = self.aggre_neigbors(torch.sum(beta.unsqueeze(2).expand_as(h_oI) * h_oI, 1)) # B x emb_dim 100 | 101 | ## learning user latent factor 102 | h_i = self.combine_mlp(torch.cat([h_iI, h_iS], dim = 1)) 103 | 104 | return h_i 105 | 106 | 107 | class _ItemModel(nn.Module): 108 | '''Item modeling to learn item latent factors. 109 | ''' 110 | def __init__(self, emb_dim, user_emb, item_emb, rate_emb): 111 | super(_ItemModel, self).__init__() 112 | self.emb_dim = emb_dim 113 | self.user_emb = user_emb 114 | self.item_emb = item_emb 115 | self.rate_emb = rate_emb 116 | 117 | self.g_u = _MultiLayerPercep(2 * self.emb_dim, self.emb_dim) 118 | 119 | self.item_users_att = _MultiLayerPercep(2 * self.emb_dim, 1) 120 | self.aggre_users = _Aggregation(self.emb_dim, self.emb_dim) 121 | 122 | self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') 123 | # used for preventing zero div error when calculating softmax score 124 | self.eps = 1e-10 125 | 126 | def forward(self, iids, i_user_pad): 127 | # user aggregation 128 | p_t = self.user_emb(i_user_pad[:,:,0]) 129 | mask_i = torch.where(i_user_pad[:,:,0] > 0, torch.tensor([1.], device=self.device), torch.tensor([0.], device=self.device)) 130 | i_user_er = self.rate_emb(i_user_pad[:,:,1]) 131 | 132 | f_jt = self.g_u(torch.cat([p_t, i_user_er], dim = 2).view(-1, 2 * self.emb_dim)).view(p_t.size()) 133 | 134 | # calculate attention scores in user aggregation 135 | q_j = mask_i.unsqueeze(2).expand_as(f_jt) * self.item_emb(iids).unsqueeze(1).expand_as(f_jt) 136 | 137 | miu = self.item_users_att(torch.cat([f_jt, q_j], dim = 2).view(-1, 2 * self.emb_dim)).view(mask_i.size()) 138 | miu = torch.exp(miu) * mask_i 139 | miu = miu / (torch.sum(miu, 1).unsqueeze(1).expand_as(miu) + self.eps) 140 | 141 | z_j = self.aggre_users(torch.sum(miu.unsqueeze(2).expand_as(f_jt) * f_jt, 1)) 142 | 143 | return z_j 144 | 145 | 146 | class GraphRec(nn.Module): 147 | '''GraphRec model proposed in the paper Graph neural network for social recommendation 148 | 149 | Args: 150 | number_users: the number of users in the dataset. 151 | number_items: the number of items in the dataset. 152 | num_rate_levels: the number of rate levels in the dataset. 153 | emb_dim: the dimension of user and item embedding (default = 64). 154 | 155 | ''' 156 | def __init__(self, num_users, num_items, num_rate_levels, emb_dim = 64): 157 | super(GraphRec, self).__init__() 158 | self.num_users = num_users 159 | self.num_items = num_items 160 | self.num_rate_levels = num_rate_levels 161 | self.emb_dim = emb_dim 162 | self.user_emb = nn.Embedding(self.num_users, self.emb_dim, padding_idx = 0) 163 | self.item_emb = nn.Embedding(self.num_items, self.emb_dim, padding_idx = 0) 164 | self.rate_emb = nn.Embedding(self.num_rate_levels, self.emb_dim, padding_idx = 0) 165 | 166 | self.user_model = _UserModel(self.emb_dim, self.user_emb, self.item_emb, self.rate_emb) 167 | 168 | self.item_model = _ItemModel(self.emb_dim, self.user_emb, self.item_emb, self.rate_emb) 169 | 170 | self.rate_pred = nn.Sequential( 171 | nn.Linear(2 * self.emb_dim, self.emb_dim, bias = True), 172 | nn.ReLU(), 173 | nn.Linear(self.emb_dim, self.emb_dim, bias = True), 174 | nn.ReLU(), 175 | nn.Linear(self.emb_dim, 1), 176 | ) 177 | 178 | 179 | def forward(self, uids, iids, u_item_pad, u_user_pad, u_user_item_pad, i_user_pad): 180 | ''' 181 | Args: 182 | uids: the user id sequences. 183 | iids: the item id sequences. 184 | u_item_pad: the padded user-item graph. 185 | u_user_pad: the padded user-user graph. 186 | u_user_item_pad: the padded user-user-item graph. 187 | i_user_pad: the padded item-user graph. 188 | 189 | Shapes: 190 | uids: (B). 191 | iids: (B). 192 | u_item_pad: (B, ItemSeqMaxLen, 2). 193 | u_user_pad: (B, UserSeqMaxLen). 194 | u_user_item_pad: (B, UserSeqMaxLen, ItemSeqMaxLen, 2). 195 | i_user_pad: (B, UserSeqMaxLen, 2). 196 | 197 | Returns: 198 | the predicted rate scores of the user to the item. 199 | ''' 200 | 201 | h_i = self.user_model(uids, u_item_pad, u_user_pad, u_user_item_pad) 202 | z_j = self.item_model(iids, i_user_pad) 203 | 204 | # make prediction 205 | r_ij = self.rate_pred(torch.cat([h_i, z_j], dim = 1)) 206 | 207 | return r_ij 208 | -------------------------------------------------------------------------------- /preprocess.py: -------------------------------------------------------------------------------- 1 | # -*- coding: utf-8 -*- 2 | """ 3 | create on Sep 24, 2019 4 | 5 | @author: wangshuo 6 | """ 7 | 8 | import random 9 | import pickle 10 | import argparse 11 | import numpy as np 12 | import pandas as pd 13 | from tqdm import tqdm 14 | from scipy.io import loadmat 15 | 16 | random.seed(1234) 17 | 18 | workdir = 'datasets/' 19 | 20 | parser = argparse.ArgumentParser() 21 | parser.add_argument('--dataset', default='Ciao', help='dataset name: Ciao/Epinions') 22 | parser.add_argument('--test_prop', default=0.1, help='the proportion of data used for test') 23 | args = parser.parse_args() 24 | 25 | # load data 26 | if args.dataset == 'Ciao': 27 | click_f = loadmat(workdir + 'Ciao/rating.mat')['rating'] 28 | trust_f = loadmat(workdir + 'Ciao/trustnetwork.mat')['trustnetwork'] 29 | elif args.dataset == 'Epinions': 30 | click_f = np.loadtxt(workdir+'Epinions/ratings_data.txt', dtype = np.int32) 31 | trust_f = np.loadtxt(workdir+'Epinions/trust_data.txt', dtype = np.int32) 32 | else: 33 | pass 34 | 35 | click_list = [] 36 | trust_list = [] 37 | 38 | u_items_list = [] 39 | u_users_list = [] 40 | u_users_items_list = [] 41 | i_users_list = [] 42 | 43 | user_count = 0 44 | item_count = 0 45 | rate_count = 0 46 | 47 | for s in click_f: 48 | uid = s[0] 49 | iid = s[1] 50 | if args.dataset == 'Ciao': 51 | label = s[3] 52 | elif args.dataset == 'Epinions': 53 | label = s[2] 54 | 55 | if uid > user_count: 56 | user_count = uid 57 | if iid > item_count: 58 | item_count = iid 59 | if label > rate_count: 60 | rate_count = label 61 | click_list.append([uid, iid, label]) 62 | 63 | pos_list = [] 64 | for i in range(len(click_list)): 65 | pos_list.append((click_list[i][0], click_list[i][1], click_list[i][2])) 66 | 67 | # remove duplicate items in pos_list because there are some cases where a user may have different rate scores on the same item. 68 | pos_list = list(set(pos_list)) 69 | 70 | # train, valid and test data split 71 | random.shuffle(pos_list) 72 | num_test = int(len(pos_list) * args.test_prop) 73 | test_set = pos_list[:num_test] 74 | valid_set = pos_list[num_test:2 * num_test] 75 | train_set = pos_list[2 * num_test:] 76 | print('Train samples: {}, Valid samples: {}, Test samples: {}'.format(len(train_set), len(valid_set), len(test_set))) 77 | 78 | with open(workdir + args.dataset + '/dataset.pkl', 'wb') as f: 79 | pickle.dump(train_set, f, pickle.HIGHEST_PROTOCOL) 80 | pickle.dump(valid_set, f, pickle.HIGHEST_PROTOCOL) 81 | pickle.dump(test_set, f, pickle.HIGHEST_PROTOCOL) 82 | 83 | 84 | train_df = pd.DataFrame(train_set, columns = ['uid', 'iid', 'label']) 85 | valid_df = pd.DataFrame(valid_set, columns = ['uid', 'iid', 'label']) 86 | test_df = pd.DataFrame(test_set, columns = ['uid', 'iid', 'label']) 87 | 88 | click_df = pd.DataFrame(click_list, columns = ['uid', 'iid', 'label']) 89 | train_df = train_df.sort_values(axis = 0, ascending = True, by = 'uid') 90 | 91 | """ 92 | u_items_list: 存储每个用户交互过的物品iid和对应的评分，没有则为[(0, 0)] 93 | """ 94 | for u in tqdm(range(user_count + 1)): 95 | hist = train_df[train_df['uid'] == u] 96 | u_items = hist['iid'].tolist() 97 | u_ratings = hist['label'].tolist() 98 | if u_items == []: 99 | u_items_list.append([(0, 0)]) 100 | else: 101 | u_items_list.append([(iid, rating) for iid, rating in zip(u_items, u_ratings)]) 102 | 103 | train_df = train_df.sort_values(axis = 0, ascending = True, by = 'iid') 104 | 105 | """ 106 | i_users_list: 存储与每个物品相关联的用户及其评分，没有则为[(0, 0)] 107 | """ 108 | for i in tqdm(range(item_count + 1)): 109 | hist = train_df[train_df['iid'] == i] 110 | i_users = hist['uid'].tolist() 111 | i_ratings = hist['label'].tolist() 112 | if i_users == []: 113 | i_users_list.append([(0, 0)]) 114 | else: 115 | i_users_list.append([(uid, rating) for uid, rating in zip(i_users, i_ratings)]) 116 | 117 | for s in trust_f: 118 | uid = s[0] 119 | fid = s[1] 120 | if uid > user_count or fid > user_count: 121 | continue 122 | trust_list.append([uid, fid]) 123 | 124 | trust_df = pd.DataFrame(trust_list, columns = ['uid', 'fid']) 125 | trust_df = trust_df.sort_values(axis = 0, ascending = True, by = 'uid') 126 | 127 | 128 | """ 129 | u_users_list: 存储每个用户互动过的用户uid； 130 | u_users_items_list: 存储用户每个朋友的物品iid列表 131 | """ 132 | for u in tqdm(range(user_count + 1)): 133 | hist = trust_df[trust_df['uid'] == u] 134 | u_users = hist['fid'].unique().tolist() 135 | if u_users == []: 136 | u_users_list.append([0]) 137 | u_users_items_list.append([[(0,0)]]) 138 | else: 139 | u_users_list.append(u_users) 140 | uu_items = [] 141 | for uid in u_users: 142 | uu_items.append(u_items_list[uid]) 143 | u_users_items_list.append(uu_items) 144 | 145 | with open(workdir + args.dataset + '/list.pkl', 'wb') as f: 146 | pickle.dump(u_items_list, f, pickle.HIGHEST_PROTOCOL) 147 | pickle.dump(u_users_list, f, pickle.HIGHEST_PROTOCOL) 148 | pickle.dump(u_users_items_list, f, pickle.HIGHEST_PROTOCOL) 149 | pickle.dump(i_users_list, f, pickle.HIGHEST_PROTOCOL) 150 | pickle.dump((user_count, item_count, rate_count), f, pickle.HIGHEST_PROTOCOL) 151 | 152 | 153 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | alabaster==0.7.12 2 | asn1crypto==0.24.0 3 | astroid==2.3.1 4 | attrs==19.2.0 5 | Babel==2.7.0 6 | backcall==0.1.0 7 | bleach==3.1.0 8 | certifi==2019.9.11 9 | cffi==1.12.3 10 | chardet==3.0.4 11 | cloudpickle==1.2.2 12 | cryptography==2.7 13 | decorator==4.4.0 14 | defusedxml==0.6.0 15 | docutils==0.15.2 16 | entrypoints==0.3 17 | idna==2.8 18 | imagesize==1.1.0 19 | ipykernel==5.1.2 20 | ipython==7.8.0 21 | ipython-genutils==0.2.0 22 | ipywidgets==7.5.1 23 | isort==4.3.21 24 | jedi==0.15.1 25 | jeepney==0.4.1 26 | Jinja2==2.10.1 27 | json5==0.8.5 28 | jsonschema==3.0.2 29 | jupyter==1.0.0 30 | jupyter-client==5.3.3 31 | jupyter-console==6.0.0 32 | jupyter-core==4.5.0 33 | jupyterlab==1.1.4 34 | jupyterlab-server==1.0.6 35 | keyring==18.0.0 36 | lazy-object-proxy==1.4.2 37 | MarkupSafe==1.1.1 38 | mccabe==0.6.1 39 | mistune==0.8.4 40 | nbconvert==5.6.0 41 | nbformat==4.4.0 42 | notebook==6.0.1 43 | numpy==1.17.2 44 | numpydoc==0.9.1 45 | packaging==19.2 46 | pandas==0.25.1 47 | pandocfilters==1.4.2 48 | parso==0.5.1 49 | pexpect==4.7.0 50 | pickleshare==0.7.5 51 | Pillow==6.2.0 52 | prometheus-client==0.7.1 53 | prompt-toolkit==2.0.10 54 | psutil==5.6.3 55 | ptyprocess==0.6.0 56 | pycodestyle==2.5.0 57 | pycparser==2.19 58 | pyflakes==2.1.1 59 | Pygments==2.4.2 60 | pylint==2.4.2 61 | pyOpenSSL==19.0.0 62 | pyparsing==2.4.2 63 | pyrsistent==0.15.4 64 | PySocks==1.7.1 65 | python-dateutil==2.8.0 66 | pytz==2019.2 67 | pyzmq==18.1.0 68 | QtAwesome==0.6.0 69 | qtconsole==4.5.5 70 | QtPy==1.9.0 71 | requests==2.22.0 72 | rope==0.14.0 73 | scipy==1.3.1 74 | SecretStorage==3.1.1 75 | Send2Trash==1.5.0 76 | six==1.12.0 77 | snowballstemmer==1.9.1 78 | Sphinx==2.2.0 79 | sphinxcontrib-applehelp==1.0.1 80 | sphinxcontrib-devhelp==1.0.1 81 | sphinxcontrib-htmlhelp==1.0.2 82 | sphinxcontrib-jsmath==1.0.1 83 | sphinxcontrib-qthelp==1.0.2 84 | sphinxcontrib-serializinghtml==1.1.3 85 | spyder==3.3.6 86 | spyder-kernels==0.5.2 87 | terminado==0.8.2 88 | testpath==0.4.2 89 | torch==1.1.0 90 | torchvision==0.3.0 91 | tornado==6.0.3 92 | tqdm==4.36.1 93 | traitlets==4.3.3 94 | urllib3==1.24.2 95 | wcwidth==0.1.7 96 | webencodings==0.5.1 97 | widgetsnbextension==3.5.1 98 | wrapt==1.11.2 99 | wurlitzer==1.0.3 100 | -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import random 3 | 4 | truncate_len = 30 5 | 6 | """ 7 | Ciao dataset info: 8 | Avg number of items rated per user: 38.3 9 | Avg number of users interacted per user: 2.7 10 | Avg number of users connected per item: 16.4 11 | """ 12 | 13 | def collate_fn(batch_data): 14 | """This function will be used to pad the graph to max length in the batch 15 | It will be used in the Dataloader 16 | """ 17 | uids, iids, labels = [], [], [] 18 | u_items, u_users, u_users_items, i_users = [], [], [], [] 19 | u_items_len, u_users_len, i_users_len = [], [], [] 20 | 21 | for data, u_items_u, u_users_u, u_users_items_u, i_users_i in batch_data: 22 | 23 | (uid, iid, label) = data 24 | uids.append(uid) 25 | iids.append(iid) 26 | labels.append(label) 27 | 28 | # user-items 29 | if len(u_items_u) <= truncate_len: 30 | u_items.append(u_items_u) 31 | else: 32 | u_items.append(random.sample(u_items_u, truncate_len)) 33 | u_items_len.append(min(len(u_items_u), truncate_len)) 34 | 35 | # user-users and user-users-items 36 | if len(u_users_u) <= truncate_len: 37 | u_users.append(u_users_u) 38 | u_u_items = [] 39 | for uui in u_users_items_u: 40 | if len(uui) < truncate_len: 41 | u_u_items.append(uui) 42 | else: 43 | u_u_items.append(random.sample(uui, truncate_len)) 44 | u_users_items.append(u_u_items) 45 | else: 46 | sample_index = random.sample(list(range(len(u_users_u))), truncate_len) 47 | u_users.append([u_users_u[si] for si in sample_index]) 48 | 49 | u_users_items_u_tr = [u_users_items_u[si] for si in sample_index] 50 | u_u_items = [] 51 | for uui in u_users_items_u_tr: 52 | if len(uui) < truncate_len: 53 | u_u_items.append(uui) 54 | else: 55 | u_u_items.append(random.sample(uui, truncate_len)) 56 | u_users_items.append(u_u_items) 57 | 58 | u_users_len.append(min(len(u_users_u), truncate_len)) 59 | 60 | # item-users 61 | if len(i_users_i) <= truncate_len: 62 | i_users.append(i_users_i) 63 | else: 64 | i_users.append(random.sample(i_users_i, truncate_len)) 65 | i_users_len.append(min(len(i_users_i), truncate_len)) 66 | 67 | batch_size = len(batch_data) 68 | 69 | # padding 70 | u_items_maxlen = max(u_items_len) 71 | u_users_maxlen = max(u_users_len) 72 | i_users_maxlen = max(i_users_len) 73 | 74 | u_item_pad = torch.zeros([batch_size, u_items_maxlen, 2], dtype=torch.long) 75 | for i, ui in enumerate(u_items): 76 | u_item_pad[i, :len(ui), :] = torch.LongTensor(ui) 77 | 78 | u_user_pad = torch.zeros([batch_size, u_users_maxlen], dtype=torch.long) 79 | for i, uu in enumerate(u_users): 80 | u_user_pad[i, :len(uu)] = torch.LongTensor(uu) 81 | 82 | u_user_item_pad = torch.zeros([batch_size, u_users_maxlen, u_items_maxlen, 2], dtype=torch.long) 83 | for i, uu_items in enumerate(u_users_items): 84 | for j, ui in enumerate(uu_items): 85 | u_user_item_pad[i, j, :len(ui), :] = torch.LongTensor(ui) 86 | 87 | i_user_pad = torch.zeros([batch_size, i_users_maxlen, 2], dtype=torch.long) 88 | for i, iu in enumerate(i_users): 89 | i_user_pad[i, :len(iu), :] = torch.LongTensor(iu) 90 | 91 | return torch.LongTensor(uids), torch.LongTensor(iids), torch.FloatTensor(labels), \ 92 | u_item_pad, u_user_pad, u_user_item_pad, i_user_pad --------------------------------------------------------------------------------