Identifying Usage Expression Sentences in Consumer Product Reviews

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 394–403, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Identifying Usage Expression Sentences in Consumer Product Reviews Shibamouli Lahiri and V. G. Vinod Vydiswaran and Rada Mihalcea University of Michigan {lahiri, vgvinodv, mihalcea}@umich.edu Abstract In this paper we introduce the problem of identifying usage expression sentences in a consumer product review. We create a human-annotated gold standard dataset of 565 reviews spanning five distinct product categories. Our dataset consists of more than 3, 000 annotated sentences. We fur- ther introduce a classification system to la- bel sentences according to whether or not they describe some “usage.” The system combines lexical, syntactic, and semantic features in a product-agnostic fashion to yield good classification performance. We show the effectiveness of our approach us- ing importance ranking of features, error analysis, and cross-product classification experiments. 1 Introduction Identification of usage expressions — phrases or sentence snippets describing product use in re- views — is an important problem in mining con- sumer product reviews. Identifying such usage expressions accurately allows us to view the re- lationship between consumers and products more clearly (e.g., by indicating how frequently a con- sumer uses a product). Further, the language and style employed in describing product use bring rel- evant and unseen aspects of the products to the fore (e.g., describing usage of a product in non- traditional and unique ways). Usage expressions can take several forms, such as which aspects of the product are used, why the product is used, where it is used, how it is used, when it is used, and so forth (c.f. Section 3 for specific examples). The product could be used by a consumer in a number of ways, sometimes in unique ways not intended for originally. Hence enumerating all possible uses of a product is com- putationally intractable. In this paper, therefore, we focus on four specific cases of product usage: why the product is used, where it is used, how it is used, and if there are any non-standard or non- traditional use (cf. Section 3). While the relationship between product usage and consumer behavior has mostly been discussed by marketing researchers and psychologists, the question of whether the phenomenon of usage has any detectable signature in terms of the language used by consumers has not been addressed thus far. In this paper, we introduce the task of iden- tifying usage expressions from consumer product reviews. In particular, we focus on classifying re- view sentences as to whether they contain a us- age expression or not. We create our own human- annotated corpus of 565 reviews on five distinct product categories containing more than 3000 sen- tences. We introduce a system that classifies sen- tences according to whether they contain a usage expression or not with 87.2% accuracy. We also show that an appropriate combination of lexical, syntactic, and semantic features performs better than individual feature categories. 2 Related Work Existing research could be organized …

Content truncated to 500 words