On the Importance of Parameter Tuning in Text Categorization

Koster, C. H. A.; Beney, Jean

doi:10.1007/978-3-540-70881-0_24

Cited by 11 publications

(7 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the text‐categorization domain, different settings are used, such as β = 16, γ = 4, and β = γ, where Moschitti () showed that β = γ is the best literature parameterization. However, Koster and Beney () found that the β parameter must be smaller (or even much smaller) than the γ parameter, in which β = 0.1 and γ = 1 were used in their study, so that a greater weight is given to negative examples.…”

Section: Important Factors Of Prfmentioning

confidence: 99%

Factors affecting rocchio‐based pseudorelevance feedback in image retrieval

Tsai

Chen

2014

Asso for Info Science & Tech

View full text Add to dashboard Cite

Pseudorelevance feedback (PRF) was proposed to solve the limitation of relevance feedback (RF), which is based on the user-in-the-loop process. In PRF, the top-k retrieved images are regarded as PRF. Although the PRF set contains noise, PRF has proven effective for automatically improving the overall retrieval result. To implement PRF, the Rocchio algorithm has been considered as a reasonable and well-established baseline. However, the performance of Rocchio-based PRF is subject to various representation choices (or factors). In this article, we examine these factors that affect the performance of Rocchio-based PRF, including image-feature representation, the number of top-ranked images, the weighting parameters of Rocchio, and similarity measure. We offer practical insights on how to optimize the performance of Rocchio-based PRF by choosing appropriate representation choices. Our extensive experiments on NUS-WIDE-LITE and Caltech 101 + Corel 5000 data sets show that the optimal feature representation is color moment + wavelet texture in terms of retrieval efficiency and effectiveness. Other representation choices are that using top-20 ranked images as pseudopositive and pseudonegative feedback sets with the equal weight (i.e., 0.5) by the correlation and cosine distance functions can produce the optimal retrieval result.

show abstract

Section: Important Factors Of Prfmentioning

confidence: 99%

Factors affecting rocchio‐based pseudorelevance feedback in image retrieval

Tsai

Chen

2014

Asso for Info Science & Tech

View full text Add to dashboard Cite

show abstract

“…In related text retrieval studies, different settings are used. For example, Moschitti [20] concluded that β ¼γ is the best literature parameterization whereas Koster and Beney [16] found that β must be smaller than γ to obtain better retrieval performance. These findings indicate that the best parameter setting should be dataset dependent.…”

Section: Further Comparisonsmentioning

confidence: 98%

The effect of low-level image features on pseudo relevance feedback

Lin

Chen

et al. 2015

Neurocomputing

View full text Add to dashboard Cite

“…Rocchio's (1971) method was one of the most famous and successful methods. Also, lots of variations with the similar idea had achieved good performances: Ide (1971) used the top irrelevant document only for feedback; Singhal et al (1997) indicated that better results could be obtained by using documents close to the query of interest only, rather than all documents; Desjardins and Godin (2000) presented the development of a different weighting formula; Nick and Themis' (2001) approach was to rank the rating of the relevance of the retrieved documents; Kim et al's (2001) study calculated the relevance degree; Azimi-Sadjadi et al (2007) study exploited relevance feedback from multiple expert users; Koster and Beney (2007) proposed the modification of parameters in Rocchio's original formula.…”

Section: The Application Of Relevance Feedbackmentioning

confidence: 99%

Construction and application of specialty-term information for document re-ranking

Chou¹,

Dai

2016

OIR

View full text Add to dashboard Cite

Purpose Conventional studies mainly classify a term’s appearance in the retrieved documents as either relevant or irrelevant for application. The purpose of this paper is to differentiate the term’s appearances in the retrieved documents in more detailed situations to generate relevance information and demonstrate the applicability of the derived information in combination with current methods of query expansion. Design/methodology/approach A method was designed first to utilize the derived information owing to term appearance differentiation within a conventional query expansion approach that has been proven as an effective technology in the enhancement of information retrieval. Then, an information retrieval system was developed to demonstrate the realization and sustain the study of the method. Formal tests were conducted to examine the distinguishing capability of the proposed information utilized in the method. Findings The experimental results show that substantial differences in performances can be achieved between the proposed method and the conventional query expansion method alone. Practical implications Since the proposed information resides at the bottom of the information hierarchy of relevance feedback, any technology regarding the application of relevance feedback information could consider the utilization of this piece of information. Originality/value The importance of the study is the disclosure of the applicability of the proposed information beyond current usage of term appearances in relevant/irrelevant documents and the initiation of a query expansion technology in the application of this information.

show abstract

On the Importance of Parameter Tuning in Text Categorization

Cited by 11 publications

References 11 publications

Factors affecting rocchio‐based pseudorelevance feedback in image retrieval

Factors affecting rocchio‐based pseudorelevance feedback in image retrieval

The effect of low-level image features on pseudo relevance feedback

Construction and application of specialty-term information for document re-ranking

Contact Info

Product

Resources

About