![](\"images/PyData.png\")
"
22 | ]
23 | },
24 | {
25 | "cell_type": "markdown",
26 | "metadata": {
27 | "slideshow": {
28 | "slide_type": "slide"
29 | }
30 | },
31 | "source": [
32 | "# About Me\n",
33 | "\n",
34 | "
"
35 | ]
36 | },
37 | {
38 | "cell_type": "markdown",
39 | "metadata": {
40 | "slideshow": {
41 | "slide_type": "slide"
42 | }
43 | },
44 | "source": [
45 | "\n",
46 | "# Delia Rusu\n",
47 | "\n",
48 | "\n",
49 | "- Chief Data Scientist @ Knowsis\n",
50 | "\n",
51 | "- PhD, Natural Language Processing and Machine Learning\n",
52 | "\n",
53 | "- Interests: unstructured data & finance"
54 | ]
55 | },
56 | {
57 | "cell_type": "markdown",
58 | "metadata": {
59 | "slideshow": {
60 | "slide_type": "slide"
61 | }
62 | },
63 | "source": [
64 | "# This Talk\n",
65 | "\n",
66 | "![](\"images/FTSE100.png\")
"
80 | ]
81 | },
82 | {
83 | "cell_type": "markdown",
84 | "metadata": {
85 | "slideshow": {
86 | "slide_type": "slide"
87 | }
88 | },
89 | "source": [
90 | "# Sources of Unstructured Data\n",
91 | "\n",
92 | "- Annual reports\n",
93 | "\n",
94 | "- Broker research\n",
95 | "\n",
96 | "- Conference call transcripts\n",
97 | "\n",
98 | "- Investor relations presentations\n",
99 | "\n",
100 | "- News and press releases\n",
101 | "\n",
102 | "- In-house content"
103 | ]
104 | },
105 | {
106 | "cell_type": "markdown",
107 | "metadata": {
108 | "slideshow": {
109 | "slide_type": "slide"
110 | }
111 | },
112 | "source": [
113 | "# What about unconventional datasets?"
114 | ]
115 | },
116 | {
117 | "cell_type": "markdown",
118 | "metadata": {
119 | "slideshow": {
120 | "slide_type": "slide"
121 | }
122 | },
123 | "source": [
124 | "# Wikipedia\n",
125 | "\n",
126 | "- 10 edits per second\n",
127 | "\n",
128 | "- the English Wikipedia \n",
129 | " - currently has over 5M articles\n",
130 | " - average of 800 new articles per day "
131 | ]
132 | },
133 | {
134 | "cell_type": "markdown",
135 | "metadata": {
136 | "slideshow": {
137 | "slide_type": "subslide"
138 | }
139 | },
140 | "source": [
141 | "# Structured Views of Wikipedia\n",
142 | "\n",
143 | "- Wikipedia article -> vectorized representation of the context\n",
144 | "\n",
145 | "- Wikipedia as a graph"
146 | ]
147 | },
148 | {
149 | "cell_type": "markdown",
150 | "metadata": {
151 | "slideshow": {
152 | "slide_type": "subslide"
153 | }
154 | },
155 | "source": [
156 | "# Word2Vec\n",
157 | "\n",
158 | "- Word2Vec model (Mikolov et al. 2013)\n",
159 | " - distributional hypothesis: words in similar contexts have similar meanings\n",
160 | " - shallow, 2-layer neural network\n",
161 | " - training objective - learn word vector representations which can predict nearby words (in context)\n",
162 | " \n",
163 | "![](\"images/SkipGramModel.png\")
"
164 | ]
165 | },
166 | {
167 | "cell_type": "markdown",
168 | "metadata": {
169 | "slideshow": {
170 | "slide_type": "subslide"
171 | }
172 | },
173 | "source": [
174 | "# Word & Article Vectors\n",
175 | "\n",
176 | "- generate context vectors for:\n",
177 | " - words which appear in Wikipedia articles\n",
178 | " - Wikipedia articles themselves (topics)\n",
179 | " \n",
180 | "- model: wiki2vec\n",
181 | " - https://github.com/idio/wiki2vec"
182 | ]
183 | },
184 | {
185 | "cell_type": "markdown",
186 | "metadata": {
187 | "slideshow": {
188 | "slide_type": "fragment"
189 | }
190 | },
191 | "source": [
192 | "**Unilever** is an Anglo-Dutch **multinational consumer goods company**"
193 | ]
194 | },
195 | {
196 | "cell_type": "markdown",
197 | "metadata": {
198 | "slideshow": {
199 | "slide_type": "fragment"
200 | }
201 | },
202 | "source": [
203 | "**Unilever_ID** is an Anglo-Dutch **multinational_consumer_goods_company_ID**"
204 | ]
205 | },
206 | {
207 | "cell_type": "markdown",
208 | "metadata": {
209 | "slideshow": {
210 | "slide_type": "subslide"
211 | }
212 | },
213 | "source": [
214 | "# Wikipedia Graph\n",
215 | "\n",
216 | "- Types of nodes:\n",
217 | " - articles, categories\n",
218 | " \n",
219 | "- Types of directed edges:\n",
220 | " - hyperlinks from article to article\n",
221 | " - infobox links from article to article \n",
222 | " - links from article to category\n",
223 | " - links from category to category"
224 | ]
225 | },
226 | {
227 | "cell_type": "markdown",
228 | "metadata": {
229 | "slideshow": {
230 | "slide_type": "subslide"
231 | }
232 | },
233 | "source": [
234 | "# Graph-Based Similarity\n",
235 | "\n",
236 | "- using the distance between nodes in the graph\n",
237 | " - in this case, $$Similarity(v_1,v_2) = 1 - Distance(v_1, v_2)$$\n",
238 | " \n",
239 | "- using random walks on graphs\n",
240 | " - Personalized PageRank (Haveliwala, 2002) is a variation of PageRank (Brin and Page, 1998)\n",
241 | " - the user query defines how important the vertex is, such that PageRank will prefer vertexes in the vicinity of the query vertex."
242 | ]
243 | },
244 | {
245 | "cell_type": "markdown",
246 | "metadata": {
247 | "slideshow": {
248 | "slide_type": "slide"
249 | }
250 | },
251 | "source": [
252 | "# FTSE 100 Index Analysis"
253 | ]
254 | },
255 | {
256 | "cell_type": "code",
257 | "execution_count": 2,
258 | "metadata": {
259 | "collapsed": false,
260 | "slideshow": {
261 | "slide_type": "fragment"
262 | }
263 | },
264 | "outputs": [
265 | {
266 | "data": {
267 | "text/html": [
268 | "![](\"images/ThisTalk.jpg\")
"
833 | ]
834 | }
835 | ],
836 | "metadata": {
837 | "celltoolbar": "Slideshow",
838 | "kernelspec": {
839 | "display_name": "Python 2",
840 | "language": "python",
841 | "name": "python2"
842 | },
843 | "language_info": {
844 | "codemirror_mode": {
845 | "name": "ipython",
846 | "version": 2
847 | },
848 | "file_extension": ".py",
849 | "mimetype": "text/x-python",
850 | "name": "python",
851 | "nbconvert_exporter": "python",
852 | "pygments_lexer": "ipython2",
853 | "version": "2.7.11"
854 | }
855 | },
856 | "nbformat": 4,
857 | "nbformat_minor": 0
858 | }
859 |
--------------------------------------------------------------------------------