Identification of Implicit Topics in Twitter Data Not Containing Explicit Search Queries

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 58–68, Dublin, Ireland, August 23-29 2014. Identification of Implicit Topics in Twitter Data Not Containing Explicit Search Queries Suzi Park Hyopil Shin Department of Linguistics, Seoul National University 1 Gwanak-ro, Gwanak-gu, Seoul, 151-745, Republic of Korea {mam3b,hpshin}@snu.ac.kr Abstract This study aims at retrieving tweets with an implicit topic, which cannot be identified by the current query-matching system employed by Twitter. Such tweets are relevant to a given query but do not explicitly contain the term. When these tweets are combined with a relevant tweet containing the overt keyword, the “serialized” tweets can be integrated into the same discourse context. To this end, features like reply relation, authorship, temporal proximity, continuation markers, and discourse markers were used to build models for detecting serialization. According to our experiments, each one of the suggested serializing methods achieves higher means of average precision rates than baselines such as the query matching model and the tf-idf weighting model, which indicates that considering an individual tweet within a discourse context is helpful in judging its relevance to a given topic. 1 Introduction 1.1 Limits of the Twitter Query-Matching Search Twitter search was not a very crucial thing in the past (Stone, 2009a), at least for users in its early stages who read and wrote tweets only within their curated timelines real-time (Dorsey, 2007; Stone, 2009b; Stone, 2009c). Users’ personal interests became one of the motivations to explore a large body of tweets only after commercial, political and academic demands, but it triggered the current extension of the Twitter search service. The domain of Twitter search was widened, for example, from tweets in the recent week to older ones (Burstein, 2013), and from accounts that have a specific term in their name or username to those that are relevant to that particular subject (Stone, 2007; Stone, 2008; Twitter, 2011; Kozak, Novermber 19, 2013). However, the standard Twitter search mechanism is based only on the presence of query terms. Even though the Twitter Search API provides many operators, the current query matching search does not guarantee retrieving a complete list of all relevant tweets.12 The 140-character limit sometimes forces a tweet not to contain a term, not because of its lack of relevance to the topic represented by the term, but due to one of the following: Reduction the query term is written in an abbreviated form or in form of Internet slang, Expansion the query term is in external text that can be expanded through other services such as Twit- Longer (http://twitlonger.com) and twtkr (http://twtkr.olleh.com), while the part exceeding 140 characters is shown only as a link on twitter.com, or Serialization the query term is contained as an antecedent in some previous tweet. This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/ 1“[T]he Search API is focused on relevance and not completeness.” 2 October 2013. Using the …

Content truncated to 500 words