Commit 612aa1b7 authored by Loïc Barrault's avatar Loïc Barrault
Browse files

added text processing (3 lectures on sentiment analysis)

parent 6af995b5
@article{Chenlo2014,
abstract = {While a number of isolated studies have analysed how different sentence features are beneficial in Sentiment Analysis, a complete picture of their effectiveness is still lacking. In this paper we extend and combine the body of empirical evidence regarding sentence subjectivity classification and sentence polarity classification, and provide a comprehensive analysis of the relative importance of each set of features using data from multiple benchmarks. To the best of our knowledge, this is the first study that evaluates a highly diversified set of sentence features for the two main sentiment classification tasks.},
author = {Jose M. Chenlo and David E. Losada},
doi = {10.1016/J.INS.2014.05.009},
issn = {0020-0255},
journal = {Information Sciences},
month = {10},
pages = {275-288},
publisher = {Elsevier},
title = {An empirical study of sentence features for subjectivity and polarity classification},
volume = {280},
url = {https://www.sciencedirect.com/science/article/pii/S0020025514005477},
year = {2014},
}
@article{Riloff2003,
author = {Ellen Riloff and Janyce Wiebe},
pages = {105-112},
title = {Learning Extraction Patterns for Subjective Expressions},
url = {https://www.aclweb.org/anthology/W03-1014/},
year = {2003},
}
@inproceedings{Yu2003,
address = {Morristown, NJ, USA},
author = {Yu, Hong and Hatzivassiloglou, Vasileios},
booktitle = {Proceedings of the 2003 conference on Empirical methods in natural language processing -},
doi = {10.3115/1119355.1119372},
file = {::},
mendeley-groups = {SentAnalysis},
pages = {129--136},
publisher = {Association for Computational Linguistics},
title = {{Towards answering opinion questions}},
url = {http://portal.acm.org/citation.cfm?doid=1119355.1119372},
volume = {10},
year = {2003}
}
@inproceedings{Hu2004,
address = {New York, New York, USA},
author = {Hu, Minqing and Liu, Bing},
booktitle = {Proceedings of the 2004 ACM SIGKDD international conference on Knowledge discovery and data mining - KDD '04},
doi = {10.1145/1014052.1014073},
isbn = {1581138889},
mendeley-groups = {SentAnalysis},
pages = {168},
publisher = {ACM Press},
title = {{Mining and summarizing customer reviews}},
url = {http://portal.acm.org/citation.cfm?doid=1014052.1014073},
year = {2004}
}
@article{Liu2005,
address = {New York, New York, USA},
author = {Liu, Bing and Hu, Minqing and Cheng, Junsheng},
doi = {10.1145/1060745.1060797},
file = {:Users/loicbarrault/Library/Application Support/Mendeley Desktop/Downloaded/Liu, Hu, Cheng - 2005 - Opinion observer analyzing and comparing opinions on the Web.pdf:pdf},
isbn = {1-59593-046-9},
journal = {Proceedings of the 14th international conference on World Wide Web},
keywords = {information extraction,opinion analysis,sentiment analysis,visualization},
mendeley-groups = {SentAnalysis},
pages = {342--351},
publisher = {ACM},
title = {{Opinion observer: analyzing and comparing opinions on the Web}},
url = {http://portal.acm.org/citation.cfm?doid=1060745.1060797},
year = {2005}
}
@inproceedings{Liu2011,
address = {San Francisco, California, USA},
author = {Liu, Bing},
booktitle = {Proceedings of AAAI},
keywords = {information extraction,opinion analysis,sentiment analysis,visualization},
title = {{Sentiment Analysis and Opinion Mining}},
url = {http://www.cs.uic.edu/~lzhang3/paper/opinion_survey.pdf},
year = {2011}
}
@misc{Bronnimann2013,
author = {Br{\'{o}}nnimann, Rebecca and Herlihy, Jane and M{\"{u}}ller, Julia and Ehlert, Ulrike},
booktitle = {The European Journal of Psychology Applied to Legal Context },
isbn = {1889-1861 UL - http://scielo.isciii.es/scielo.php?script=sci{\_}arttext{\&}pid=S1889-18612013000100005{\&}nrm=iso},
mendeley-groups = {SentAnalysis},
pages = {97--121},
publisher = {scieloes },
title = {{Do testimonies of traumatic events differ depending on the interviewer? }},
volume = {5 },
year = {2013}
}
This diff is collapsed.
This diff is collapsed.
% !TEX root = text_processing.tex
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{}
\vfill
\centering
\Huge{\edinred{Sentiment Analysis\\Lexicon based approaches}}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: 2 main approaches}
\only<1>{
\begin{itemize}
\item Lexicon based
\begin{itemize}
\item Binary
\item Gradable
\end{itemize}
\item Corpus based
\begin{itemize}
\item Naive Bayes
\item Deep Learning
\end{itemize}
\end{itemize}
}
\only<2>{
\begin{itemize}
\item Lexicon based
\begin{itemize}
\item \textbf{Binary}
\item Gradable
\end{itemize}
\item Corpus based
\begin{itemize}
\item Naive Bayes
\item Deep Learning
\end{itemize}
\end{itemize}
}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based}
\begin{block}{Principle}
\begin{itemize}
\item Use a lexicon of opinion/emotion words \ra\ words with \myemph{polarity}
\item[\ra] e.g. good, bad, horrible, great, etc.
\end{itemize}
\end{block}
Rule-based sentiment classifier at \textbf{sentence} or \textbf{document} level
\begin{enumerate}
\item<2-> Rule-based \textbf{subjectivity classifier}
\begin{itemize}
\item text is \myemph{subjective} if it has $n$ words from the emotion lexicon ($n$ is fixed by expert)
\item \myemph{objective} otherwise
\end{itemize}
\item<3-> Rule-based \textbf{sentiment classifier}
\begin{itemize}
\item applied on \myemph{subjective} text only
\item \textbf{count} the number of positive and negative word/phrases in the text
\item text is
\begin{itemize}
\item \red{\bf negative} if more negative than positive
\item \green{\bf positive} if more positive than negative
\item \orange{\bf neutral} otherwise
\end{itemize}
\end{itemize}
\end{enumerate}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: binary}
Rule-based \textbf{sentiment classifier} at \textbf{feature} level
\begin{itemize}
\item Assume \myemph{feature} can be identified in a previous step \ra\ battery, phone, screen
\item Identify \myemph{emotion} associated with those \myemph{features}
\item count \red{\bf negative} and \green{\bf positive} emotion words/phrases rom the lexicon
\item feature is
\begin{itemize}
\item \red{\bf negative} if more negative than positive
\item \green{\bf positive} if more positive than negative
\item \orange{\bf neutral} otherwise
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: binary}
Rule-based \textbf{sentiment classifier} at \textbf{feature} level
\begin{itemize}
\item<+-> Simple approach:
\begin{itemize}
\item \textbf{input}: a pair $(f,S)$ where $f$ is a product feature and $S$ is a sentence containing $f$
\item \textbf{output}: a label in either \red{\bf negative}, \green{\bf positive} or \orange{\bf neutral}.
\end{itemize}
\item<+-> Protocol: consider $S = w_1,..., w_N$ the sentence containing $f$, with $N$ its length
\begin{enumerate}
\item select the \myemph{emotion} words $w_i$ in $S$
\item assign \textbf{orientations} to each of these words $w_i$
\begin{itemize}
\item \red{\bf negative} \ra\ -1
\item \green{\bf positive} \ra\ +1
\item \orange{\bf neutral} \ra\ 0
\end{itemize}
\item \textbf{sum up} the orientation and \textbf{assign} a label to $(f,S)$ accordingly
\end{enumerate}
\item<+-> more advanced strategies:
\begin{itemize}
\item split the sentence in part using \textbf{discourse connectives/markers}
\item focus on markers that may introduce a change in the sentiment
\item[\ra] "but", "except that"
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: binary: caveats}
Certain words have context-independent orientations, e.g. “great”. \\
Other emotion words have \myemph{context-dependent orientations}, e.g.
\begin{itemize}
\item "\green{small power consumption}" is positive but \red{small capacity} is negative
\item "\red{consume valuable resources}" is negative but \green{consume disgusting waste} is positive
\end{itemize}
One has to deal with \myemph{negation}, e.g.:
\begin{itemize}
\item \red{not great} is negative but \green{not bad} is positive
\end{itemize}
One has to deal with \myemph{intensifiers}:
\begin{itemize}
\item \green{very good} is more positive than \green{good}
\item \red{extremely boring} is more negative than \red{boring} or \red{very boring}
\end{itemize}
\vspace{.5cm}
\ra\ need a more \textbf{fine-grained sentiment information} in lexicon and add \textbf{additional rules}.
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: 2 main approaches}
\begin{itemize}
\item Lexicon based
\begin{itemize}
\item {\bf \color{lightgray} Binary}
\item \textbf{Gradable}
\end{itemize}
\item Corpus based
\begin{itemize}
\item Naive Bayes
\item Deep Learning
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
\begin{block}{Principle}
\begin{itemize}
\item Use of \textbf{ranges of sentiment} instead of a \textbf{binary system}
\item deal with \myemph{intensifiers} like: absolutely, utterly, completely, totally, nearly, virtually, essentially, mainly, almost, ...
\item[\ra] e.g.: \myemph{absolutely} \red{\bf awful}
\end{itemize}
\end{block}
\textbf{Grade} adverbs like:
\begin{itemize}
\item Very, little, dreadfully, extremely, fairly, hugely, immensely, intensely, rather, reasonably, slightly, unusually, ...
\item[\ra] e.g.: \green{\bf immensely beautiful}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
Rule-based \textbf{gradable sentiment classifier}
\begin{itemize}
\item \textbf{Classify} general \myemph{valence} of a text based on \textbf{the level of emotional content}
\item level of emotional content given by:
\begin{enumerate}
\item<2-> the \textbf{lexicon}: word list with pre-assigned emotional weights
\begin{itemize}
\item \red{{\bf negative dimension}: $C_{neg} \in {-5,...,-1}$}
\item \green{{\bf positive dimension}: $C_{pos} \in {+1,...,+5}$}
\item \orange{\bf neutral (0)} is ignored
\end{itemize}
\item<3-> additional \textbf{general rules}: word list with pre-assigned emotional weights
\begin{itemize}
\item \myemph{negation rule}
\item \myemph{capitalization rule}
\item \myemph{intensifier rule}
\item \myemph{diminisher rule}
\item \myemph{exclamation rule}
\item \myemph{emoticon rule}
\end{itemize}
\end{enumerate}
\end{itemize}
\only<2->{
\begin{textblock*}{50mm}[0,0](120mm,15mm)
\includegraphics[width=.5\textwidth]{word_lexicon_gradable}
\end{textblock*}}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
\begin{enumerate}
\setcounter{enumi}{1}
\item additional \textbf{general rules}: word list with pre-assigned emotional weights
\end{enumerate}
\begin{itemize}
\item \myemph{\bf negation rule}: \textbf{inverse} and \textbf{reduce} the weight when the word "not" is in the neighbourhood
\begin{itemize}
\item Ex.: "I am not good today" and in the lexicon: \green{\bf emotion(good) = 3}
\item \orange{reduce} the value by 1 and \blue{inverse} the polarity
\item[\ra] \red{\bf new emotion(good)} = $(+3 ~ \orange{-1}) ~ \blue{ * -1}$ = \red{\bf -2}
\end{itemize}
\item[]
\item \myemph{\bf capitalization rule}: \textbf{strengthen} the weight when written in \textbf{capitals}
\begin{itemize}
\item Ex.: "I am GOOD today"
\item[\ra] \green{\bf new emotion(good)} = $(+3 ~ \liumcyan{+1})$ = \green{\bf +4}
\item Ex.: "I am AWFUL today" and in the lexicon: \red{\bf emotion(awful) = -4}
\item[\ra] \red{\bf new emotion(awful)} = $(-4 ~ \liumcyan{-1})$ = \red{\bf -5}
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
\begin{enumerate}
\setcounter{enumi}{1}
\item additional \textbf{general rules}: word list with pre-assigned emotional weights
\end{enumerate}
\begin{itemize}
\item \myemph{\bf intensifier rule}: \textbf{strengthen} the weight of the modified word
\begin{itemize}
\item Needs a list of intensifiers with their weights
\item[\ra] similar to the word lexicon: weight(very) = 1, weight(extremely) = 2, etc.
\item the weight is \green{\bf added} to \green{\bf positive} terms
\item the weight is \red{\bf substracted} to \red{\bf negative} terms
\item Ex.: "I am feeling very good"
\item[\ra] \green{\bf new emotion(good)} = $(+3 ~ \liumcyan{+1})$ = \green{\bf +4}
\item Ex.: "This was an extremely boring game" and in the lexicon: \red{\bf emotion(boring) = -3}
\item[\ra] \red{\bf new emotion(boring)} = $(-3 ~ \liumcyan{-2})$ = \red{\bf +5}
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
\begin{enumerate}
\setcounter{enumi}{1}
\item additional \textbf{general rules}: word list with pre-assigned emotional weights
\end{enumerate}
\begin{itemize}
\item \myemph{\bf diminisher rule}: \textbf{weaken} the weight of the modified word
\begin{itemize}
\item Needs a list of diminishers with their weights
\item[\ra] similar to the word lexicon: weight(somewhat) = 1, weight(slightly) = 1, etc.
\item the weight is \red{\bf substracted} to \green{\bf positive} terms
\item the weight is \green{\bf added} to \red{\bf negative} terms
\item Ex.: "I am somewhat good"
\item[\ra] \green{\bf new emotion(good)} = $(+3 ~ \liumcyan{-1})$ = \green{\bf +2}
\item Ex.: "This was an slightly boring game" and in the lexicon: \red{\bf emotion(boring) = -3}
\item[\ra] \red{\bf new emotion(boring)} = $(-3 ~ \liumcyan{+1})$ = \red{\bf -2}
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
\begin{enumerate}
\setcounter{enumi}{1}
\item additional \textbf{general rules}: word list with pre-assigned emotional weights
\end{enumerate}
\begin{itemize}
\item \myemph{\bf exclamation rule}: \textbf{strengthen} the weight of the modified word (similar to \myemph{intensifiers})
\begin{itemize}
\item Needs a list of exclamations with their weights
\item[\ra] similar to the word lexicon: weight(!!!) = 2, weight(!!!!!!!!!) = 3, etc.
\item the weight is \green{\bf added} to \green{\bf positive} terms
\item the weight is \red{\bf substracted} to \red{\bf negative} terms
\item Ex.: "Great show!!!" and in the lexicon: \green{\bf emotion(great) = +3}
\item[\ra] \green{\bf new emotion(great)} = $(+3 ~ \liumcyan{+2})$ = \green{\bf +5}
% \item Ex.: "This was an slightly boring game" and in the lexicon: \red{\bf emotion(boring) = -3}
% \item[\ra] \red{\bf new emotion(boring)} = $(-3 ~ \liumcyan{+1})$ = \red{\bf -2}
\end{itemize}
\item[]
\item \myemph{\bf emoticon rule}: provide a weight to each emoticon (similar to words in \myemph{lexicon})
\begin{itemize}
\item Needs a list of emoticons with their weights
\item \green{emotion({\DejaSans}) = +2}, \red{emotion({\DejaSans}) = -2}
% %😐😁😂😃😇😉😈😋😍😱
\item Ex1.: "I can't believe this product {\DejaSans}" \ra\ \green{\bf emotion(Ex1.)} = \green{\bf +2}
\item Ex2.: "I can't believe this product {\DejaSans}" \ra\ \red{\bf emotion(Ex2.)} = \red{\bf -2}
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: gradable}
The \myemph{valence} of the text is the sum of the weights of the emotional words
Consequently:
\begin{itemize}
\item if $|C_{pos}| < |C_{neg}|$ then \red{{\bf emotion(text) = negative}}
\item if $|C_{pos}| > |C_{neg}|$ then \green{{\bf emotion(text) = positive}}
\item if $|C_{pos}| = |C_{neg}|$ then \orange{{\bf emotion(text) = neutral}}
\end{itemize}
\begin{itemize}
\item[]
\item Ex1.: text = "He is brilliant but boring", emotion(brilliant) = +2, emotion(boring) = -3
\begin{itemize}
\item[\ra] \myemph{valence(text)} = +2 - 3 = -1 \ra\ \red{{\bf negative}}
\end{itemize}
\item Ex2.: text = "I am not good today", emotion(good) = +2, \myemph{negation rule}
\begin{itemize}
\item[\ra] \myemph{valence(text)} = +2 * -1 = -2 \ra\ \red{{\bf negative}}
\end{itemize}
\item Home exercises:
\begin{itemize}
\item Ex3. "I am not GOOD today"
\item Ex4. "I am so surprised by this product!!! {\DejaSans}"
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based}
\begin{itemize}
\item \myemph{Advantages}
\begin{itemize}
\item Works effectively with different texts: forums, blogs, etc.
\item Language independent \ra\ only need a lexicon with emotion weights
\item No training data required
\item Easily extendible: simply add new entries in the dictionary
\end{itemize}
\item \myemph{Disadvantages}
\begin{itemize}
\item The lexicon of emotion word is a resource made by experts
\begin{itemize}
\item costly to create and maintain
\end{itemize}
\item Needs frequent updates to incorporate new words/abbreviations
\item Not robust to spelling errors
\item[\ra] A dataset from MySpace has 95\% of comments containing at least one spelling error
\item Static resource
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based}
\textbf{How to obtain lexica of emotion words?}
\begin{enumerate}
\item Collect relevant words/phrases expressing sentiment
\item Determine the emotion of these subjective texts
\item[\ra] Can be done \textbf{manually}
\begin{itemize}
\item word list with pre-assigned emotional weights
\end{itemize}
\item[\ra] Can be done \textbf{semi-automatically}
\begin{itemize}
\item require a dictionary of \textbf{seed emotion words}
\item \myemph{dictionary-based}
\begin{itemize}
\item find synonyms/antonyms using linguistic resources like e.g. WordNet
\item[\ra] \url{https://wordnet.princeton.edu} and \url{http://projects.illc.uva.nl/EuroWordNet}
\item[\ra] http://globalwordnet.org/resources/wordnets-in-the-world/
\end{itemize}
\item \myemph{corpus-based}
\begin{itemize}
\item find synonyms/antonyms in corpora
\end{itemize}
\end{itemize}
\end{enumerate}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based}
\textbf{What do lexica of emotion words contain?}
\begin{itemize}
\item \myemph{Adjectives}
\begin{itemize}
\item \green{\bf positive}: honest, important, mature, large, patient, ...
\item \red{\bf negative}: harmful, hypocritical, inefficient, insecure, ...
\end{itemize}
\item \myemph{Verbs}
\begin{itemize}
\item \green{\bf positive}: praise, love, ...
\item \red{\bf negative}: blame, criticize, ...
\end{itemize}
\item \myemph{Nouns}
\begin{itemize}
\item \green{\bf positive}: pleasure, enjoyment, ...
\item \red{\bf negative}: pain, criticism, ...
\end{itemize}
\item \myemph{Phrases}: for \textbf{collocations}, also an alternative to \myemph{intensifiers}
\begin{itemize}
\item \green{\bf positive}: "very efficient", "low cost", ...
\item \red{\bf negative}: "many problems", "lot of bugs", ...
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sentiment Analysis: lexicon based: manually created resources}
Manually created resources by experts\\
\begin{itemize}
\item \textbf{SentiWordNet} database: \url{http://ontotext.fbk.eu/sentiwn.html}
\begin{itemize}
\item \textbf{Wordnet}: words grouped in sets of synonyms (\textbf{synsets})
\item with semantic relations between them: synonyms, antonyms, hypernyms, etc.
\item[\ra] sentiment score added \red{\bf negative}, \green{\bf positive} and \orange{\bf neutral}
\end{itemize}
\item \textbf{Linguistic Inquiry and Word Count (LIWC) lexicon}
\begin{itemize}
\item made by psychologists
\item words with several \textbf{emotional and other dimensions}
\end{itemize}
\item \textbf{General Inquirer}
\begin{itemize}
\item terms with various types of \red{\bf negative} or \green{\bf positive} semantic orientation
\item[\ra] \url{http://www.wjh.harvard.edu/~inquirer/homecat.htm}
\end{itemize}
\end{itemize}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{SentiWordNet}
\includegraphics[width=.9\textwidth]{sentiwn_ex}\\