cogs 105 this week Types of Research • Philosophical / theoretical BIG DA T A • Experimental • Observational • Computational • Cognitive engineering today: latent semantic analysis Types of Research Experimental vs. Observational involves direct intervention is avoided • Philosophical / theoretical intervention (or not possible) • Experimental • Observational • Computational • Cognitive engineering E.g., setup experimental task in laboratory for babies Deb Roy, MIT
Experimental vs. Observational Experimental vs. Observational dependent variable outcome variable causal correlational (you measure) (variable of interest) inferences often inferences are acceptable preferred independent variable predictors and covariates (you control) (to predict / explain outcome) Enhanced social familiarity Outcome: Extent of play is related to increased Enhanced social familiarity Predictor: Depth of social familiarity play engagement. DV: Extent of play causes increased play engagement Covariates: Time of day, recent food, etc. IV: Depth of social familiarity Big Data Example • Remember, “big data” is a general term that connotes a • Facebook’s controversial study. trend to utilize large and unseemly data sets to render new insights. • Studies using big data are primarily observational in nature. (Correlational studies with lots of data.) Experimental evidence of massive-scale emotional Significance contagion through social networks • Big data studies can sometimes be experimental Adam D. I. Kramer a,1 , Jamie E. Guillory b,2 , and Jeffrey T. Hancock b,c We show, via a massive ( N = 689,003) experiment on Facebook, though. (Use of technology to setup experimental a Core Data Science Team, Facebook, Inc., Menlo Park, CA 94025; and Departments of b Communication and c Information Science, Cornell University, Ithaca, that emotional states can be transferred to others via emotional conditions and collect lots of data.) NY 14853 contagion, leading people to experience the same emotions Edited by Susan T. Fiske, Princeton University, Princeton, NJ, and approved March 25, 2014 (received for review October 23, 2013) without their awareness. We provide experimental evidence Emotional states can be transferred to others via emotional demonstrated that ( i ) emotional contagion occurs via text-based contagion, leading people to experience the same emotions that emotional contagion occurs without direct interaction be- computer-mediated communication (7); ( ii ) contagion of psy- • Also big data can be used to build tools for without their awareness. Emotional contagion is well established chological and physiological qualities has been suggested based tween people (exposure to a friend expressing an emotion is in laboratory experiments, with people transferring positive and on correlational data for social networks generally (7, 8); and experimental research. negative emotions to others. Data from a large real-world social ( iii ) people ’ s emotional expressions on Facebook predict friends ’ sufficient), and in the complete absence of nonverbal cues. network, collected over a 20-y period suggests that longer-lasting emotional expressions, even days later (7) (although some shared moods (e.g., depression, happiness) can be transferred through experiences may in fact last several days). To date, however, there networks [Fowler JH, Christakis NA (2008) BMJ 337:a2338], al- is no experimental evidence that emotions or moods are contagious though the results are controversial. In an experiment with people in the absence of direct interaction between experiencer and target. who use Facebook, we test whether emotional contagion occurs On Facebook, people frequently express emotions, which are outside of in-person interaction between individuals by reducing later seen by their friends via Facebook ’ s “ News Feed ” product the amount of emotional content in the News Feed. When positive (8). Because people s friends frequently produce much more
cogs 105 this week Linguistic Tools • Big data can also help us render new tools — for BIG DA T A example, the development of semantic models. • Latent semantic analysis (LSA). • Uses massive amounts of text to build a model that allows us to compare words to each other in terms of their “meaning.” • Thursday: LIWC today: latent semantic analysis Starting Point Mapping Meaning • LSA goes from a huge amount of text data, to a distilled representation of word meaning in the form of a vector space or “map.” • In this space, words do not have “meaning” all on their own; their meanings are derived from their relationships to other words. dog cat car break brake work
How LSA Works: Map Description How LSA Works: Juicing Description “massive text info” LSA “word “massive “word meaning” LSA text meaning” info” How LSA Works: Almost There How LSA Works: Almost There Step 1: Word-by-Document Matrix Files / documents “corpus” dog cat LSA car break Words brake work cells represent how often a word occurs in each file “dog” (represented by grayscale)
The Problem A Simple Motivation… • “dog” may rarely or even never occur in the same • The cells in a word-by-document matrix are mostly document as either “parrot” or “pencil.” empty; this creates great difficulties in relating word meaning. • However, both “parrot” and “dog” may occur with similar words: “breathe, eat, drink, noise, interact, • Sometimes called “data sparsity” problem. owner,” etc. • LSA is a statistical techniques that acts like • LSA is able to extract these relationships — and so “squeezing the sponge” or “drawing the map” by it would tell us, in our map of meaning, that “dog” extracting the major trends/relationships among and “parrot” are more similar than “dog” and words in the matrix . “pencil.” Finally… How LSA Works: Almost There Step 2: LSA space is a lower dimensional matrix Files / documents Dimensions Dimensions singular value decomposition LSA Words LSA the dimensions (…this is our are now the space “map” or the “dog” in which words live “dog” “dog” “juice”…) and can be related
If dimensions happen to be really small (1, 2, Why “LSA”? smaller angle, cosine would be closer to 1 or 3) we can visualize them like this: • Latent = “existing but not yet developed or cat manifest; hidden.” dog cos(angle) • Semantic = “of or related to meaning.” • Analysis = …analysis. bark angle fly LSA airplane bigger angle, cosine closer to 0 “Meaning” So How Do I LSA? • Modern cognitive science methods now allow us to • Do I have to crunch all the numbers? “quantify meaning” in this way. • It’s actually pretty easy to do it. If you want sample • Philosophers have spent millennia talking about code, I can show you how to build an LSA model in meaning; there is still endless debate about no more than 10 lines of code in MATLAB, Python, meaning. or R. • However, LSA, as a model of meaning , can grade • However, for the purposes of this class and explore papers, pass the MCAT, work with educational LSA, we will use an amazing online tool… technologies, and many more.
Matrix Comparison lsa.colorado.edu Sentences / Passages? Running Some Comparisons • What about sentences? What if we want to compare larger blocks of text?
What’s It Good For? Running Some Comparisons • Tons of stuff! E.g.: • Experimental design (e.g., controlling for word similarity in an RT task) • Observational designs (e.g., comparing semantic similarity between conversation partners; e.g., Dale & Duran, 2008) • Search engine and document indexing • Educational technologies (e.g., artificial tutors) Limitations • LSA suffers from some problems. • It can’t handle syntax. • E.g., these words have the “same meaning” • The dog ate my homework • The homework ate my dog (?)
Limitations • It does not do well with homonymy (“same word, different meanings”). • E.g., “cream in your coffee” and “cream you at hockey” have different “creams” in them. • LSA treats them as one word. Limitations • It does not do well with antonymy (opposites). • Love and hate occur in overlapping descriptive contexts, but they are quite different in meaning. • LSA often treats antonyms as similar in meaning (could this make sense sometimes?)
Despite Limitations… Next Time • We’ll compare quantitative and qualitative approaches with LIWC, in the context of Big Data. • Lab this week: Neurosynth.
Recommend
More recommend