• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
TweetSpector: Entity-based retrieval of Tweets [Demo]
 

TweetSpector: Entity-based retrieval of Tweets [Demo]

on

  • 427 views

This is a demonstration, which will be presented by Surender Reddy Yerva during the 35th Annual SIGIR (Special Interest Group on Information Retrieval) Conference, taking place in Portland, Oregon, ...

This is a demonstration, which will be presented by Surender Reddy Yerva during the 35th Annual SIGIR (Special Interest Group on Information Retrieval) Conference, taking place in Portland, Oregon, USA from August 12-16, 2012.

Extended Abstract:
People readily express their opinions about the various products, companies, TV shows etc., on Twitter. These
tweet messages are thus a rich source of information that can be exploited to understand the sentiments about the concerned products or services. Retrieving the tweets related to given entities is however a challenging task as their names are often (deliberately) ambiguous, e.g. Apple, Blackberry, Friends, etc. Nevertheless, identifying the relevant entities is an essential rst step to develop reliable sentiment analysis techniques that is not considered in existing systems, for example TweetFeel, TwitterSentiment.
While there is a number of techniques for identifying namedentities in unstructured text, they are often not directly applicable in this case, as tweet messages are very short (maximal 140 characters). This demonstrator introduces TweetSpector, a tool that addresses this retrieval task and enables to link tweet messages to a given entity. Our retrieval methods rely on classi cation techniques that exploit our concise descriptions of entity-relevant information, also called entity pro les.

Statistics

Views

Total Views
427
Views on SlideShare
400
Embed Views
27

Actions

Likes
0
Downloads
0
Comments
0

2 Embeds 27

http://www.planet-data.eu 18
http://planet-data.eu 9

Accessibility

Categories

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    TweetSpector: Entity-based retrieval of Tweets [Demo] TweetSpector: Entity-based retrieval of Tweets [Demo] Document Transcript

    • TweetSpector: Entity-based retrieval of Tweets Surender Reddy Yerva, Zoltán Miklós, Flavia Grosan, Alexandru Tandrau, Karl Aberer Swiss Federal Institute of Technology (EPFL) Lausanne, Switzerland {surenderreddy.yerva,zoltan.miklos,flavia.grosan,alexandru.tandrau,karl.aberer}@epfl.chCategories and Subject DescriptorsH.3.1 [Information Systems Applications]: Content Anal-ysis and Indexing; H.3.5 [Information Systems Applica-tions]: Online Information ServicesKeywordsEntity, Disambiguation, Profiles, Twitter1. EXTENDED ABSTRACT People readily express their opinions about the variousproducts, companies, TV shows etc., on Twitter1 . Thesetweet messages are thus a rich source of information that canbe exploited to understand the sentiments about the con-cerned products or services. Retrieving the tweets relatedto given entities is however a challenging task as their namesare often (deliberately) ambiguous, e.g. Apple, Blackberry,Friends, etc. Nevertheless, identifying the relevant entitiesis an essential first step to develop reliable sentiment analy-sis techniques that is not considered in existing systems, forexample TweetFeel2 , TwitterSentiment3 . While there is a number of techniques for identifying named Figure 1: TweetSpector: Various Featuresentities in unstructured text, they are often not directly ap-plicable in this case, as tweet messages are very short (max- -Tweet Classification: TweetSpector displays in real-timeimal 140 characters). This demonstrator introduces Tweet- the classification results (see Figure 1). For example, aSpector, a tool that addresses this retrieval task and enables stream of tweets is displayed and it is indicated whetherto link tweet messages to a given entity. Our retrieval meth- or not the messages shall be related to the company Ap-ods rely on classification techniques that exploit our concise ple Inc.. The classification techniques are widely extendeddescriptions of entity-relevant information, also called entity versions of our earlier work [1].profiles. -User Feedback: The users can indicate whether the pro- The demonstrator presents the following features of Tweet- posed classification is correct or not. This feedback is takenSpector: into account by the algorithms. TweetSpector can also take -Entity Profile Creation: TweetSpector supports auto- human input through crowdsourcing (through an interfacematic profile creation, where we apply named-entity recog- to Amazon Mechanical Turk).nition, NLTK, wordnet and Web data extraction techniques -Dashboard: TweetSpector can display performance met-to construct profiles for an entity, given a relevant Web- rics and statistical information on a dashboard related topage. TweetSpector also enables manual profile construc- the entity.tion, where users can construct arbitrary entity profiles,as well as manual and automatic updates for initially con- 2. ACKNOWLEDGEMENTSstructed profiles (thus the profiles are dynamic). The profiles This work was partly funded by the NisB project (FP7-can also be visualized using Word Clouds. ICT-256955) and the European Commission in the Planet-1 Data NoE (contract nr. 257641). http://www.twitter.com2 http://www.tweetfeel.com3 http://twittersentiment.appspot.com 3. REFERENCES [1] Surender Reddy Yerva, Zolt´n Mikl´s, and Karl a o Aberer. Entity-based Classification of TwitterCopyright is held by the author/owner(s). Messages. International Journal of Computer Science &SIGIR’12, August 12–16, 2012, Portland, Oregon, USA.ACM 978-1-4503-1472-5/12/08. Applications, 9(1):88–115, 2012.