Abstract: With the rapid development of e-commerce and significant growth of user-generated review data, product search has been faced with more and more challenges over the years. Product review data are useful as they provide a more trusted and subjective way of description. However, as for review data, traditional retrieval models usually suffer from data sparsity and term weight uniformity. To address these problems, this paper incorporates word semantic association into existing retrieval models, proposes a query expansion method to correct existing retrieval models, and implements the method under a high-performance retrieval framework. The experimental results on several types of product data show that the method achieves better precision as well as higher quality, which will be potentially helpful for product retrieval.

Key words: word association, review retrieval, semantic expansion

摘要： 随着电子商务的快速发展和用户在线评论数据的迅速增加，产品评论检索面临更多的挑战。一方面，产品评论从更为主观的角度为产品的特性提供描述；另一方面，产品评论的数据特性要求对传统检索方法进行相应的修正，以解决数据稀疏和词项权重单一等问题。在产品评论检索的任务下，引入词项相关度的概念，针对传统检索方法主题词项稀疏和词项权重缺少先验的问题，进行基于语义的查询扩展。同时，将词项相关度融入到一个高性能的检索框架中。一系列评测实验表明，该方法可以提高产品评论检索的准确率与质量，更好地提升评论的参考价值。

关键词: 词项相关度, 评论检索, 语义扩展