Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Intent Attribute-Aware Text Matching in Searching

Mingzhe Li, Xiuying Chen, Jing Xiang, Qishen Zhang, Changsheng Ma, Chenchen Dai, Jinxiong Chang, Zhongyi Liu, Guannan Zhang

TL;DR

The paper tackles the challenge of text matching in search when both queries and items carry rich attributes. It introduces Multi-Intent Attribute-Aware Matching (MIM), a three-component framework comprising an attribute-aware encoder, multi-intent modeling, and intent-aware matching, trained with distribution, KL-divergence, and intent-masking losses to encourage diverse yet aligned intents. Empirical results show consistent offline gains across Amazon and Alipay datasets and positive online impact in a live Alipay app, demonstrating improved relevance and user experience. The work highlights the practical value of explicitly modeling cross-side attribute intents and lays groundwork for extending to multimodal attribute-aware matching.

Abstract

Text matching systems have become a fundamental service in most searching platforms. For instance, they are responsible for matching user queries to relevant candidate items, or rewriting the user-input query to a pre-selected high-performing one for a better search experience. In practice, both the queries and items often contain multiple attributes, such as the category of the item and the location mentioned in the query, which represent condensed key information that is helpful for matching. However, most of the existing works downplay the effectiveness of attributes by integrating them into text representations as supplementary information. Hence, in this work, we focus on exploring the relationship between the attributes from two sides. Since attributes from two ends are often not aligned in terms of number and type, we propose to exploit the benefit of attributes by multiple-intent modeling. The intents extracted from attributes summarize the diverse needs of queries and provide rich content of items, which are more refined and abstract, and can be aligned for paired inputs. Concretely, we propose a multi-intent attribute-aware matching model (MIM), which consists of three main components: attribute-aware encoder, multi-intent modeling, and intent-aware matching. In the attribute-aware encoder, the text and attributes are weighted and processed through a scaled attention mechanism with regard to the attributes' importance. Afterward, the multi-intent modeling extracts intents from two ends and aligns them. Herein, we come up with a distribution loss to ensure the learned intents are diverse but concentrated, and a kullback-leibler divergence loss that aligns the learned intents. Finally, in the intent-aware matching, the intents are evaluated by a self-supervised masking task, and then incorporated to output the final matching result.

Multi-Intent Attribute-Aware Text Matching in Searching

TL;DR

The paper tackles the challenge of text matching in search when both queries and items carry rich attributes. It introduces Multi-Intent Attribute-Aware Matching (MIM), a three-component framework comprising an attribute-aware encoder, multi-intent modeling, and intent-aware matching, trained with distribution, KL-divergence, and intent-masking losses to encourage diverse yet aligned intents. Empirical results show consistent offline gains across Amazon and Alipay datasets and positive online impact in a live Alipay app, demonstrating improved relevance and user experience. The work highlights the practical value of explicitly modeling cross-side attribute intents and lays groundwork for extending to multimodal attribute-aware matching.

Abstract

Text matching systems have become a fundamental service in most searching platforms. For instance, they are responsible for matching user queries to relevant candidate items, or rewriting the user-input query to a pre-selected high-performing one for a better search experience. In practice, both the queries and items often contain multiple attributes, such as the category of the item and the location mentioned in the query, which represent condensed key information that is helpful for matching. However, most of the existing works downplay the effectiveness of attributes by integrating them into text representations as supplementary information. Hence, in this work, we focus on exploring the relationship between the attributes from two sides. Since attributes from two ends are often not aligned in terms of number and type, we propose to exploit the benefit of attributes by multiple-intent modeling. The intents extracted from attributes summarize the diverse needs of queries and provide rich content of items, which are more refined and abstract, and can be aligned for paired inputs. Concretely, we propose a multi-intent attribute-aware matching model (MIM), which consists of three main components: attribute-aware encoder, multi-intent modeling, and intent-aware matching. In the attribute-aware encoder, the text and attributes are weighted and processed through a scaled attention mechanism with regard to the attributes' importance. Afterward, the multi-intent modeling extracts intents from two ends and aligns them. Herein, we come up with a distribution loss to ensure the learned intents are diverse but concentrated, and a kullback-leibler divergence loss that aligns the learned intents. Finally, in the intent-aware matching, the intents are evaluated by a self-supervised masking task, and then incorporated to output the final matching result.
Paper Structure (19 sections, 8 equations, 4 figures, 5 tables)

This paper contains 19 sections, 8 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model
  4. Problem Formulation
  5. Attribute-aware Encoder
  6. Multi-Intent Modeling
  7. Intent-aware Matching
  8. Experiments
  9. Datasets
  10. Baselines.
  11. Implementation Details
  12. Evaluation Metrics
  13. Offline Performance
  14. Online Experiments
  15. ANALYSIS AND DISCUSSION
  16. ...and 4 more sections

Figures (4)

  • Figure 2: The application of our model in the query-item matching setting. The model identifies multiple intents on "cost-effective", "skin clean", and "light color", based on which it conducts query-item matching to return relevant items to the users.
  • Figure 3: Overview of MIM, which consists of three parts: (1) Attribute-aware Encoder obtains vector representations for the queries and attributes based on cross-attention and attribute-weighting; (2) Multi-Intent Modeling extracts multiple diverse intents related to the queries with two devised losses; (3) Intent-Aware Matching incorporates information from queries and intents to output the final matching result.
  • Figure 4: (a) Relationship between number of intent and Accuracy, AUC and F1 score on Alipay test dataset. (b) Performance of MIM after removing different attributes in Alipay dataset.
  • Figure 5: Visualization of intent distribution with attributes, where ten cases are sampled from the test set randomly. The red nodes, blue nodes, and green nodes represent the intents, attributes, and queries, respectively.