Jason S Kessler icwsm09

Citation

Targeting Sentiment Expressions through Supervised Ranking of Linguistic Configurations, Jason S. Kessler, Nicolas Nicolov, ICWSM 09

Online version

An online pdf version is here[1]

Summary

The focus of this paper is on the problem of finding the targets of individual expressions of sentiment. To do this task, the author first build an annotated data containing 194 blog entries that contained a paragraph or longer evaluative passage about a car or a digital camera.

Each entity, Each mention of an entity, Contextual sentiment toward entities as well as their referent objects, Part of and feature-of relations, Sentiment expressions and their modifiers and their mentions they target, Immediate opinion holders, are annotated in the data.

After having this data, they use 4 approaches to solve the problem. 1) Proximity (namely the nearest) 2) Heuristic syntax (baseline when using parsing results) 3)Bloom (pattern list from Bloom Garg and Argamon) 4) RankSVM (learn a model that ranks mentions occurring in the same sentence as a sentiment expression by using feature like Lexical Distance, Lexical Path, POS Relation etc.)

Evaluation

The author report that the RankSVM perform better than the other three methods. And the Bloom results is surprisingly poor in recall.

Discussion

There are some interesting numbers in Section 6: 1) 51% of targets appeared to the left of the sentiment expression and 49% appeared to the right. 2) While the median number of tokens between a sentiment expression and its target is 2, the mean is 6.21. Making matters worse, 41% of sentiment expressions have a mention they do not target at least as close to them as a target. 3) 30% have one that is closer. 4) 91% of targets are in the same sentence as their sentiment expression. 5) 70% of noun phrase targets are common noun phrases, 14% are pronouns, and 16% are proper names.