[Paper Review] Intent term selection and refinement in e-commerce queries

Saurav Manchanda, Mohit Sharma+1 more|arXiv (Cornell University)|Aug 22, 2019

Web Data Mining and Analysis21 references4 citations

TL;DR

This paper proposes context-aware term weighting and query refinement methods for e-commerce search using query reformulation logs from Walmart. By leveraging RNN-based intent encoding and contextual term representation, the approach improves relevance for rare queries by identifying intent-critical terms and suggesting vocabulary-gap bridging terms, outperforming non-contextual baselines in MRR and precision metrics.

ABSTRACT

In e-commerce, a user tends to search for the desired product by issuing a query to the search engine and examining the retrieved results. If the search engine was successful in correctly understanding the user's query, it will return results that correspond to the products whose attributes match the terms in the query that are representative of the query's product intent. However, the search engine may fail to retrieve results that satisfy the query's product intent and thus degrading user experience due to different issues in query processing: (i) when multiple terms are present in a query it may fail to determine the relevant terms that are representative of the query's product intent, and (ii) it may suffer from vocabulary gap between the terms in the query and the product's description, i.e., terms used in the query are semantically similar but different from the terms in the product description. Hence, identifying the terms that describe the query's product intent and predicting additional terms that describe the query's product intent better than the existing query terms to the search engine is an essential task in e-commerce search. In this paper, we leverage the historical query reformulation logs of a major e-commerce retailer to develop distant-supervised approaches to solve both these problems. Our approaches exploit the fact that the significance of a term is dependent upon the context (other terms in the neighborhood) in which it is used in order to learn the importance of the term towards the query's product intent. We show that identifying and emphasizing the terms that define the query's product intent leads to a 3% improvement in ranking. Moreover, for the tasks of identifying the important terms in a query and for predicting the additional terms that represent product intent, experiments illustrate that our approaches outperform the non-contextual baselines.

Motivation & Objective

• To improve e-commerce search relevance for rare or cold-start queries where historical engagement data is limited.
• To identify which terms in a query most accurately represent the user's product intent, especially amid noisy or ambiguous terms.
• To bridge the vocabulary gap between user queries and product catalog terms by suggesting more relevant, contextually appropriate terms.
• To develop a generalizable method that leverages query reformulation patterns to infer intent without relying on explicit annotations.

Proposed method

• Uses historical query reformulation logs from Walmart to train a distant-supervised model for intent detection.
• Employs a recurrent neural network (RNN) encoder to model the contextual representation of query terms, capturing how term significance varies by surrounding entities.
• Introduces a term-weighting model (CTW) that assigns higher weights to terms that best express the query’s product intent, based on context-aware embeddings.
• Develops a query refinement model (CQR) using an intent encoder and multilabel classifier to predict relevant terms not present in the original query.
• Combines term importance estimation with vocabulary gap resolution by predicting terms that better match product catalog language.
• Evaluates models using MRR and precision metrics on a held-out set of rare queries, comparing against TF-IDF, FTW, VPCG, and VG baselines.

Experimental results

Research questions

• RQ1 How can we identify the most relevant terms in a search query that express the true product intent, especially when the query contains ambiguous or noisy terms?
• RQ2 To what extent does incorporating contextual information from neighboring terms improve term weighting accuracy compared to non-contextual methods?
• RQ3 Can query reformulation patterns be leveraged to suggest new, more effective terms that bridge the vocabulary gap between user queries and product catalog descriptions?
• RQ4 How effective are context-aware models in improving search performance for rare queries with limited historical engagement?

Key findings

• The context-aware term weighting model (CTW) outperforms non-contextual baselines like TF-IDF, FTW, VPCG, and VG in ranking relevance, as measured by higher MRR and precision.
• For the query 'battery night light with timer', CTW correctly assigns the highest weights to 'night' and 'light', while baselines incorrectly prioritize 'timer' or 'battery'.
• The query refinement model (CQR) successfully predicts contextually relevant terms such as 'water spray nozzle' for 'orbit red garden hose water nozzle', avoiding spurious associations like 'gum' or 'spearmint'.
• In the case of 'auto seat cover wonder woman', CTW correctly identifies 'auto', 'seat', and 'cover' as key terms, while baselines fail to disambiguate the intent.
• The CQR model avoids generating irrelevant terms by leveraging the intent encoder to understand the product type, unlike the FQR baseline which produces incorrect suggestions due to lack of context.
• The approach demonstrates strong performance on rare queries, where traditional methods fail due to lack of historical engagement data, confirming its value for cold-start scenarios.