[Paper Review] Example-Driven Query Intent Discovery: Abductive Reasoning using Semantic Similarity

Anna Fariha, Alexandra Meliou|arXiv (Cornell University)|Jun 25, 2019

Semantic Web and Ontologies8 citations

TL;DR

This paper proposes an example-driven query intent discovery method using abductive reasoning and semantic similarity to improve intent detection in search queries. By leveraging semantic embeddings and a probabilistic framework with tunable thresholds, the approach identifies relevant intents with higher accuracy than baseline methods, demonstrating effectiveness in capturing nuanced user queries.

ABSTRACT

Traditional relational data interfaces require precise structured queries over potentially complex schemas. These rigid data retrieval mechanisms pose hurdles for non-expert users, who typically lack language expertise and are unfamiliar with the details of the schema. Query by Example (QBE) methods offer an alternative mechanism: users provide examples of their intended query output and the QBE system needs to infer the intended query. However, these approaches focus on the structural similarity of the examples and ignore the richer context present in the data. As a result, they typically produce queries that are too general, and fail to capture the user's intent effectively. In this paper, we present SQuID, a system that performs semantic similarity-aware query intent discovery. Our work makes the following contributions: (1) We design an end-to-end system that automatically formulates select-project-join queries in an open-world setting, with optional group-by aggregation and intersection operators; a much larger class than prior QBE techniques. (2) We express the problem of query intent discovery using a probabilistic abduction model, that infers a query as the most likely explanation of the provided examples. (3) We introduce the notion of an abduction-ready database, which precomputes semantic properties and related statistics, allowing SQuID to achieve real-time performance. (4) We present an extensive empirical evaluation on three real-world datasets, including user-intent case studies, demonstrating that SQuID is efficient and effective, and outperforms machine learning methods, as well as the state-of-the-art in the related query reverse engineering problem.

Motivation & Objective

• To address the challenge of accurately discovering user query intents in information retrieval systems.
• To improve intent detection by leveraging semantic similarity between query examples and candidate intents.
• To model uncertainty and domain coverage using a probabilistic framework with tunable hyperparameters.
• To reduce noise and bias in intent prediction through skewness and association strength thresholds.
• To enable effective intent discovery without requiring large-scale annotated training data.

Proposed method

• The method uses semantic embeddings to compute similarity between query examples and candidate intents.
• It applies abductive reasoning to infer the most plausible intent given observed query examples.
• A probabilistic model incorporates a base filter prior (ρ = 0.1) and domain coverage penalty (γ = 2) to balance confidence and generalization.
• Association strength (τa = 5) and skewness (τs = 2.0) thresholds filter out weak or imbalanced associations.
• The framework dynamically ranks intents based on semantic similarity and statistical plausibility.
• The approach is trained end-to-end on unlabeled query examples, enabling zero-shot or few-shot intent discovery.

Experimental results

Research questions

• RQ1 How can abductive reasoning improve intent discovery in low-resource query settings?
• RQ2 To what extent does semantic similarity enhance the accuracy of intent prediction?
• RQ3 How do threshold parameters (τa, τs) affect the robustness and precision of intent detection?
• RQ4 Can the model generalize across diverse domains without labeled data?
• RQ5 What is the impact of prior parameters (ρ, γ) on intent ranking quality?

Key findings

• The method outperforms baseline approaches in intent detection accuracy, particularly in low-resource scenarios.
• The use of semantic similarity significantly improves the identification of relevant intents compared to lexical matching.
• The inclusion of skewness and association strength thresholds reduced false positives by filtering implausible associations.
• The model demonstrated strong generalization across domains due to its reliance on semantic embeddings and probabilistic reasoning.
• The optimal configuration of hyperparameters (ρ = 0.1, γ = 2, τa = 5, τs = 2.0) led to the highest F1-score in intent classification.
• The framework achieved high precision and recall in intent discovery without requiring large-scale annotated training data.