Layfield, C.; Azzopardi, J,; Staff, C.: Experiments with document retrieval from small text collections using Latent Semantic Analysis or term similarity with query coordination and automatic relevance feedback (2017)
0.02
0.015495917 = product of:
0.04648775 = sum of:
0.04648775 = weight(_text_:search in 3478) [ClassicSimilarity], result of:
0.04648775 = score(doc=3478,freq=6.0), product of:
0.1747324 = queryWeight, product of:
3.475677 = idf(docFreq=3718, maxDocs=44218)
0.05027291 = queryNorm
0.2660511 = fieldWeight in 3478, product of:
2.4494898 = tf(freq=6.0), with freq of:
6.0 = termFreq=6.0
3.475677 = idf(docFreq=3718, maxDocs=44218)
0.03125 = fieldNorm(doc=3478)
0.33333334 = coord(1/3)
- Abstract
- One of the problems faced by users of databases containing textual documents is the difficulty in retrieving relevant results due to the diverse vocabulary used in queries and contained in relevant documents, especially when there are only a small number of relevant documents. This problem is known as the Vocabulary Gap. The PIKES team have constructed a small test collection of 331 articles extracted from a blog and a Gold Standard for 35 queries selected from the blog's search log so the results of different approaches to semantic search can be compared. So far, prior approaches include recognising Named Entities in documents and queries, and relations including temporal relations, and represent them as `semantic layers' in a retrieval system index. In this work, we take two different approaches that do not involve Named Entity Recognition. In the first approach, we process an unannotated version of the PIKES document collection using Latent Semantic Analysis and use a combination of query coordination and automatic relevance feedback with which we outperform prior work. However, this approach is highly dependent on the underlying collection, and is not necessarily scalable to massive collections. In our second approach, we use an LSA Model generated by SEMILAR from a Wikipedia dump to generate a Term Similarity Matrix (TSM). We automatically expand the queries in the PIKES test collection with related terms from the TSM and submit them to a term-by-document matrix derived by indexing the PIKES collection using the Vector Space Model. Coupled with a combination of query coordination and automatic relevance feedback we also outperform prior work with this approach. The advantage of the second approach is that it is independent of the underlying document collection.
- Source
- Semantic keyword-based search on structured data sources: COST Action IC1302. Second International KEYSTONE Conference, IKC 2016, Cluj-Napoca, Romania, September 8-9, 2016, Revised Selected Papers. Eds.: A. Calì, A. et al
