Better than the real thing?

Abstract

We present a novel approach to pseudo-feedback-based ad hoc retrieval that uses language models induced from both documents and clusters. First, we treat the pseudo-feedback documents produced in response to the original query as a set of pseudo-queries that themselves can serve as input to the retrieval process. Observing that the documents returned in response to the pseudo-queries can then act as pseudo-queries for subsequent rounds, we arrive at a formulation of pseudo-query-based retrieval as an iterative process. Experiments show that several concrete instantiations of this idea, when applied in conjunction with techniques designed to heighten precision, yield performance results rivaling those of a number of previously-proposed algorithms, including the standard language-modeling approach. The use of cluster-based language models is a key contributing factor to our algorithms' success.

Incorporating Semantic Word Representations into Query Expansion for Microblog Information Retrieval

“…Shen et al [26] proposed to add user clicks as implicit feedback for secondary sorting of the document, and obtained better results. Kurland et al [12] also proposed to repeat query expansion by iterations, but experiments show that the method suffered from problems such as query theme drifts.…”

A few examples go a long way

“…Kurland et al [12] provide an iterative "pseudo-query" generation technique to uncover aspects of a query, using cluster-based language models.…”

“…Next, we combine the selected query terms with the terms from the original query; this is also done in the original query expansion papers (see, e.g., [19]) and in query modeling methods based on language models (see, e.g., [12]) and prevents the topic to shift (too far) away from the original user information need. We use Eq.…”

“…The reason for this is as follows. "Aspect recall" is an important cause of failure of current IR systems [5], one that tends to be exacerbated by today's query expansion approaches: key aspects of the user's information need may be completely missing from the pool of top-ranked documents, as this pool is usually query-biased and (to keep precision reasonable) often small, and, hence, tends to only reflect aspects covered by the original query itself [12]. In a scenario such as ours, where a user provides a query plus sample documents, we expect the sample documents to provide important aspects not covered by the query.…”

Incorporating Non-Relevance Information in the Estimation of Query Models

“…As to the query model θ Q , we adopt the common approach to linearly interpolate the initial query with an expanded part (Balog et al, 2008;Kurland et al, 2005;Rocchio, 1971;Zhai and Lafferty, 2001):…”