Using Sequence Kernels to identify Opinion Entities in Urdu

Proceedings of the Fifteenth Conference on Computational Natural Language Learning, pages 58–67, Portland, Oregon, USA, 23–24 June 2011. c⃝2011 Association for Computational Linguistics Using Sequence Kernels to identify Opinion Entities in Urdu Smruthi Mukund† and Debanjan Ghosh* Rohini K Srihari †SUNY at Buffalo, NY smukund@buffalo.edu *Thomson Reuters Corporate R&D debanjan.ghosh@thomsonreuters.com SUNY at Buffalo, NY rohini@cedar.buffalo.edu Abstract Automatic extraction of opinion holders and targets (together referred to as opinion entities) is an important subtask of senti- ment analysis. In this work, we attempt to accurately extract opinion entities from Urdu newswire. Due to the lack of re- sources required for training role labelers and dependency parsers (as in English) for Urdu, a more robust approach based on (i) generating candidate word sequences corresponding to opinion entities, and (ii) subsequently disambiguating these se- quences as opinion holders or targets is presented. Detecting the boundaries of such candidate sequences in Urdu is very differ- ent than in English since in Urdu, grammatical categories such as tense, gender and case are captured in word inflections. In this work, we exploit the morphological inflections associated with nouns and verbs to correctly identify sequence boundaries. Different levels of information that capture context are encoded to train standard linear and se- quence kernels. To this end the best per- formance obtained for opinion entity detection for Urdu sentiment analysis is 58.06% F-Score using sequence kernels and 61.55% F-Score using a combination of sequence and linear kernels. 1 Introduction Performing sentiment analysis on newswire da- ta facilitates the development of systems capable of answering perspective questions like “How did people react to the latest presidential speech?” and “Does General Musharraf support the Indo-Pak peace treaty?”. The components involved in de- veloping such systems require accurate identifica- tion of opinion expressions and opinion entities. Several of the approaches proposed in the literature to automatically extract the opinion entities rely on the use of thematic role labels and dependency parsers to provide new lexical features for opinion words (Bethard et al., 2004). Semantic roles (SRL) also help to mark the semantic constituents (agent, theme, proposition) of a sentence. Such features are extremely valuable for a task like opinion en- tity detection. English is a privileged language when it comes to the availability of resources needed to contribute features for opinion entity detection. There are other widely spoken, resource poor languages, which are still in the infantile stage of automatic natural language processing (NLP). Urdu is one such language. The main objective of our research is to provide a solution for opinion entity detection in the Urdu language. Despite Urdu lacking NLP resources required to contribute features similar to what works for the English language, the perform- ance of our approach is comparable with English for this task (compared with the work of Weigand and Klalow, 2010 ~ 62.61% F1). The morphologi- cal richness of the Urdu language enables us to extract features based on noun and verb inflections that effectively contribute to the opinion entity ex- traction task. Most importantly, …

Content truncated to 500 words