[Paper Review] Knowledge-based Query Expansion in Real-Time Microblog Search

Runwei Qiang, Feifan Fan+2 more|arXiv (Cornell University)|Mar 13, 2015

Web Data Mining and Analysis24 references4 citations

TL;DR

This paper proposes a knowledge-based query expansion method for real-time microblog search that leverages Freebase to infer contextually relevant terms and improve query understanding. By integrating temporal evidence into a language modeling framework, the approach boosts recent, relevant tweets, significantly outperforming baseline methods on TREC Twitter corpora with optimized interpolation and decay parameters.

ABSTRACT

Since the length of microblog texts, such as tweets, is strictly limited to 140 characters, traditional Information Retrieval techniques suffer from the vocabulary mismatch problem severely and cannot yield good performance in the context of microblogosphere. To address this critical challenge, in this paper, we propose a new language modeling approach for microblog retrieval by inferring various types of context information. In particular, we expand the query using knowledge terms derived from Freebase so that the expanded one can better reflect users' search intent. Besides, in order to further satisfy users' real-time information need, we incorporate temporal evidences into the expansion method, which can boost recent tweets in the retrieval results with respect to a given topic. Experimental results on two official TREC Twitter corpora demonstrate the significant superiority of our approach over baseline methods.

Motivation & Objective

• Address the severe vocabulary mismatch problem in microblog search due to short, informal tweets.
• Improve query understanding by expanding queries with knowledge terms from Freebase, including aliases, notable features, and descriptions.
• Incorporate temporal evidence into query expansion to favor recent, relevant tweets in real-time retrieval.
• Balance relevance and recency by modeling document recency through time priors in a language modeling framework.
• Evaluate the method on standard TREC Twitter test collections to demonstrate superiority over state-of-the-art baselines.

Proposed method

• Use Freebase to identify relevant concepts matching the query, extracting terms from properties like alias, notable_for, and description.
• Construct a knowledge query by selecting top terms from Freebase using an association-based method with a time prior.
• Apply an exponential decay function to temporal evidence, where the rate parameter r controls the emphasis on recent terms in pseudo-relevance documents.
• Interpolate the original query with the knowledge query using a coefficient α to balance original and expanded terms.
• Apply model-based feedback (SMM) to further refine the query using top-ranked documents, controlled by a second interpolation coefficient β.
• Integrate time priors into the language model to favor documents published recently, enhancing real-time performance.

Experimental results

Research questions

• RQ1 Can knowledge-based query expansion using Freebase improve retrieval performance in microblog search by mitigating vocabulary mismatch?
• RQ2 How does incorporating temporal evidence affect the ranking of recent, relevant tweets in real-time microblog retrieval?
• RQ3 What is the optimal balance between original query, knowledge-expanded terms, and model-based feedback in a multi-stage query expansion framework?
• RQ4 How do different parameter settings (e.g., r, α, β) affect the trade-off between precision at top ranks and overall relevance (MAP)?
• RQ5 Can knowledge terms from Freebase help reduce topic drift in query expansion compared to traditional PRF methods?

Key findings

• The proposed QEFB method with r = 0.1 achieved the best balance of high and stable performance across P@N (1 ≤ N ≤ 30) and MAP scores, outperforming QEFBNT and other r settings.
• Setting α = 0.5 in the first-stage expansion yielded optimal performance, outperforming both pure original query (α=0) and pure knowledge query (α=1) approaches.
• The second-stage feedback with β = 0.6 provided the best trade-off, improving P@30 while maintaining strong MAP scores, though performance declined when β > 0.3.
• Using r = 0.5 improved top-5 precision (P@5) significantly, but led to lower MAP and degraded performance at N ≥ 10, indicating overemphasis on recency.
• The method significantly outperformed baseline approaches on two official TREC Twitter test collections, demonstrating clear superiority in real-time microblog retrieval.
• The knowledge query expansion effectively captured relevant context, such as associating 'water shortage' with 'drought' and 'Africa', reflecting real-time user information needs.