Academic
Publications
Diversifying search results

Diversifying search results,10.1145/1498759.1498766,Rakesh Agrawal,Sreenivas Gollapudi,Alan Halverson,Samuel Ieong

Diversifying search results   (Citations: 67)
BibTex | RIS | RefWorks Download
We study the problem of answering ambiguous web queries in a setting where there exists a taxonomy of information, and that both queries and documents may belong to more than one category according to this taxonomy. We present a systematic approach to diversifying results that aims to minimize the risk of dissatisfaction of the average user. We propose an algorithm that well approximates this objective in general, and is provably optimal for a natural special case. Furthermore, we generalize several classical IR metrics, including NDCG, MRR, and MAP, to explicitly account for the value of diversification. We demonstrate empirically that our algorithm scores higher in these generalized metrics compared to results produced by commercial search engines.
Conference: Web Search and Data Mining - WSDM , pp. 5-14, 2009
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
    • ...More recently, work by Clarke et al. [7,9], Agrawal et al. [1], and Chapelle et al. [5] propose weighted linear combinations of measures computed with respect to the individual subtopics, and these proposals form the focus of our paper...
    • ...Agrawal et al. [1], and Chapelle et al. [5] all model diversity by assigning a probability pi ,1 ≤ i ≤ M , to each subtopic, indicating the probability that a user entering the query is seeking information related to subtopic i...
    • ...Agrawal et al. [1] consider the case when queries are strictly ambiguous and a user is never interested in more than one interpretation of the query...
    • ...On the other hand, as we illustrate in Section 2.3, the intent-aware measures of Agrawal et al. [1] do not measure novelty because they are built upon traditional effectiveness measures...

    Charles L. A. Clarkeet al. A comparative analysis of cascade measures for novelty and diversity

    • ...users who find at least one relevant result) [27, 20, 2, 11]...
    • ...evaluation model in Eq. (2) reduces to the intent-aware evaluation measures proposed in [2]...
    • ...But for measures that account for query ambiguity, such as the measures defined above and their special cases proposed in [2], the situation is less clear...
    • ...We measure performance using the dynamic variants of Prec@10, AP@10, DCG@10, and nDCG@10 as defined in Section 3, which reduce to the intent-aware metrics of Agrawal et al. [2] for static ranking...

    Christina Brandtet al. Dynamic ranked retrieval

    • ...How can then the search engine best answer such queries? The approach taken in [5] is to diversify the search results so that the probability that an average user will find at least one relevant document in the retrieved result is maximized...

    Rakesh Agrawal. Data Externality

    • ...While addressing relevance is comparatively straightforward, and has been heavily studied in both database and information retrieval areas, diversity is a more difficult problem to solve [11], [12]...
    • ...Agrawal et al. [11] describe a greedy algorithm which relies on an objective function, computed based on a probabilistic model, that admits a sub-modularity structure and a taxonomy to compute diverse results...

    Marcos R. Vieiraet al. On query result diversification

    • ...Experiments show that our algorithm increases the expected performance of the top 10 results by 130% compared to a commercial search engine and 51% over a state-of-the-art diversification algorithm [1] in certain cases, while still performing well on traditional metrics designed under single relevant document assumptions...
    • ...Agrawal et al. introduce a model similar to ours in [1], where their objective is to maximize the probability an average user finds at least one useful result...
    • ...Classic ranked retrieval metrics such as NDCG, MRR, and MAP have been augmented [6, 1] to take user intent into account...
    • ...In this section we briefly go over the work on search diversification by Agrawal et al. [1] to better understand when prior work may perform sub-optimally, and how our approach may overcome such scenarios...
    • ...For each metric, we compare our Diversity-IQ algorithm against the state-of-the-art IA-Select algorithm presented by Agrawal et al. [1] as well as the original ranking returned by a commercial web search engine (SE)...
    • ...early, we consider the “intent aware” Mean Reciprocal Rank (MRR-IA) metric defined in [1]...

    Michael J. Welchet al. Search result diversity for informational queries

Sort by: