A New Relevance Feedback Algorithm Based on Vector Space Basis Change

Salton and Buckley’s Landmark Research in Experimental Text Information Retrieval

The idea of relevance feedback (RF) is to involve the user in the retrieval process to improve the final result set by reformulating the query. The most commonly used methods in RF aim to rewrite the user query. In the vector space model, RF is usually undertaken by re-weighting the query terms without any modification in the vector space basis. In this paper we propose a RF method based on vector space basis change without any modification on the query term weights. The aim of our method is to find a basis which gives a better representation of the documents such that the relevant documents are gathered and the irrelevant ones are kept away from the relevant documents.

In vector space models, traditional relevance feedback techniques, which utilize the terms in the relevant documents to enrich the user's initial query, is an effective method to improve retrieval performance. However, in this process, it also brings some non-relevance terms in the relevant documents in the new query. The number of non-relevance terms will increase according to the repeat of feedback process; it will damage the retrieval performance finally. This paper introduces a SVM Based method for relevance feedback. We train a classifier on the feedback documents and classify the rest of the documents. Thus, in the result list, the relevant documents are in front of the non-relevant documents. The new approach avoids modifying the query via text classification algorithm in the relevance feedback process, and it is a new direction for the relevance feedback techniques. Experiments with TREC dataset demonstrate the effectiveness of this method.

Content-based image retrieval (CBIR) has been receiving intensive research attention for many applications. In order to provide the users with more precise retrieval results, relevance feedback (RF) methods have been incorporated into CBIR which take the user's feedbacks into account. In general, explicit RF methods demand too much user effort while implicit RF methods suffer from lower retrieval accuracy. As such, we propose a hybrid RF method, adaptive-weight hybrid relevance feedback (AHRF) for content-based image retrieval. AHRF integrates explicit user grading and implicit user browsing histories to build a user preference model. The model is refined iteratively and used to train a preference classifier for the users. Moreover, an adaptive-weight mechanism is proposed to achieve a personalized preference model. Our proposed method is tested on a subset of the Corel Database and the experimental results reveal that AHRF can achieve good retrieval precision with less user effort.

Pseudo relevance feedback (PRF) enhances the retrieval performance of the relevance feedback. Pseudo relevance feedback assumes that the k highest-ranking documents in the first retrieval are relevant and extract query expansion from them. Rocchio algorithm is a classical algorithm for implementing relevance feedback into vector space models. The Rocchio algorithm forms a new query moves toward the centroid of the relevant documents and keeps away from centroid of the irrelevant documents. However, in the relevance feedback method, irrelevant documents are ignored. In this paper, we conduct a method for pseudo irrelevance feedback (PIRF) documents components that effectively applied to the Rocchio algorithm. Documents with a high ranking outside of k relevant documents and those documents dissimilar to any k relevant documents can extract good query expansion if the documents are applied as irrelevant documents. The Rocchio algorithm uses PRF as a component of relevant documents and this research method for irrelevant documents as a component of irrelevant documents denoted by Roc PRF PIRF (filter). Experiment on CISI dataset show that Roc PRF PIRF (filter) improved performance by testing several variations the number of irrelevant documents compared to the standard Rocchio algorithm and Rocchio algorithm with irrelevant documents but without proposed method).

In the task of information filtering, the profile of the user's interest and preference is the key to the performance of the system. In traditional method, the profile usually represented as a set of features in the vector space model, but this kind of profile could not be satisfied by the user for lack of the negative information. This paper proposes on approach to construct the user's profile based on negative relevance feedback and Reinforcement Learning(RL) This approach uses not only positive feedbacknnbut also negative feedback, and suggests relevance feedback method with reinforcement learning. The proposed method improves the performance of filtering system by eliminating the documents that users don't prefer. We carried filtering about four topics so as to compare the proposed method to Rocchio and Widrow-Hoff(WH), representative relevance feedback methods. The experimental result shows that the performance of the case, where only positive relevance feedback is used, is 3 to 6% better than that of Rocchio and 2 to 3% better than that of Widrow-Hoff. And the performance of the case, where both positive and negative relevance feedback are used, is 6 to 10% better than that of Rocchio and 2 to 8% better than that of Widrow-Hoff.

Negative relevance feedback is a special case of relevance feedback where we do not have any positive example; this often happens when the topic is difficult and the search results are poor. Although in principle any standard relevance feedback technique can be applied to negative relevance feedback, it may not perform well due to the lack of positive examples. In this paper, we conduct a systematic study of methods for negative relevance feedback. We compare a set of representative negative feedback methods, covering vector-space models and language models, as well as several special heuristics for negative feedback. Evaluating negative feedback methods requires a test set with sufficient difficult topics, but there are not many naturally difficult topics in the existing test collections. We use two sampling strategies to adapt a test collection with easy topics to evaluate negative feedback. Experiment results on several TREC collections show that language model based negative feedback methods are generally more effective than those based on vector-space models, and using multiple negative models is an effective heuristic for negative feedback. Our results also show that it is feasible to adapt test collections with easy topics for evaluating negative feedback methods through sampling.

In this paper, we present a new method for query reweighting to deal with document retrieval. The proposed method uses genetic algorithms to reweight a user's query vector, based on the user's relevance feedback, to improve the performance of document retrieval systems. It encodes a user's query vector into chromosomes and searches for the optimal weights of query terms for retrieving documents by genetic algorithms. After the best chromosome is found, the proposed method decodes the chromosome into the user's query vector for dealing with document retrieval. The proposed query reweighting method can find the best weights of query terms in the user's query vector, based on the user's relevance feedback. It can increase the precision rate and the recall rate of the document retrieval system for dealing with document retrieval

Machine learning techniques have been widely used in almost every area of arts, science, and technology for the last two decades. Document analysis and query expansion also use machine learning techniques at a broad scale for information retrieval tasks. The state-of-the-art models like the Bo1 model, Bo2 model, KL divergence model, and chi-square model are probabilistic, and they work on DFR-based retrieval models. These models are much focused on term frequency and do not care about the semantic relationship among the terms. The proposed model applies the semantic method to find the semantic similarity among the terms to expand the query. The proposed method uses the relevance feedback method that selects a user-assisted most relevant document from top “k” initially retrieved documents and then applies deep neural network technique to select the most informative terms related to original query terms. The results are evaluated at FIRE 2011 ad hoc English test collection. The mean average precision of the proposed method is 0.3568. The proposed method also compares the state-of-the-art models. The proposed model observed 19.77% and 8.05% improvement on the mean average precision (MAP) parameter with respect to the original query and Bo1 model, respectively.

As an effective technique for improving retrieval effectiveness, relevance feedback (RF) has been widely studied in both monolingual and cross-language information retrieval (CLIR) settings. The studies of RF in CLIR have been focused on query expansion (QE), in which queries are reformulated before and/or after they are translated. However, RF in CLIR actually not only can help select better query terms, but also can enhance query translation by adjusting translation probabilities and even resolve some out-of-vocabulary terms. In this paper, we propose a novel RF method called translation enhancement (TE), which uses the extracted translation relationships from relevant documents to revise the translation probabilities of query terms and to identify extra translation alternatives if available so that the translated queries are more tuned to the current search. We studied TE using pseudo relevance feedback (PRF) and interactive relevance feedback (IRF). Our results show that TE can significantly improve CLIR with both types of RF methods, and that the improvement is comparable to that of QE. More importantly, the effects of TE and QE are complementary. Their integration can produce further improvement, and makes CLIR more robust for a variety of queries.

Relevance Feedback (RF) is a technique allowing to enrich an initial query according to the user feedback in order to get results closer to the user's information need. This paper presents a new RF method for keyword queries (content queries). It is based on the re-weighting of the original query terms plus the addition of new query terms from the content of elements jugded as relevant or non-relevant by the user. This RF method is integrated in our search engine, Garnata, and evaluated with the INEX 2007 collection.

Intelligent techniques for effective information retrieval

Enhancing query translation with relevance feedback in translingual information retrieval

Vector space model (VSM) is an Information Retrieval (IR) system model that represents query and documents as n-dimension vector. GVSM is an expansion from VSM that represents the documents base on similarity value between query and minterm vector space of documents collection. Minterm vector is defined by the term in query. Therefore, in retrieving a document can be done base on word meaning inside the query. On the contrary, a document can consist the same information semantically. LSI is a method implemented in IR system to retrieve document base on overall meaning of users’ query input from a document, not based on each word translation. LSI uses a matrix algebra technique namely Singular Value Decomposition (SVD). This study discusses the performance of VSM, GVSM and LSI that are implemented on IR to retrieve Indonesian sentences document of .pdf, .doc and .docx extension type files, by using Nazief and Adriani stemming algorithm. Each method implemented either by thread or no-thread. Thread is implemented in preprocessing process in reading each document from document collection and stemming process either for query or documents. The quality of information retrieval performance is evaluated based-on time response, values of recall, precision, and F-measure were measured. The results show that for each method, the fastest execution time is .docx extension type file followed by .doc and .pdf. For the same document collection, the results show that time response for LSI is more faster, followed by GVSM then VSM. The average of recall value for VSM, GVSM and LSI are 82.86 %, 89.68 % and 84.93 % respectively. The average of precision value for VSM, GVSM and LSI are 64.08 %, 67.51 % and 62.08 % respectively. The average of F-measure value for VSM, GVSM and LSI are 71.95 %, 76.63 % and 71.02 % respectively. Implementation of multithread for preprocessing for VSM, GVSM, and LSI can increase average time response required is about 30.422%, 26.282%, and 31.821% respectively.

Relevance feedback (RF) is a popular reranking technique, which aims at improving the performance of image retrieval by taking the user's opinions into account. In this paper, we introduce a new RF method, named circular relevance feedback (CRF), to enhance the behavior of remote sensing image retrieval (RSIR). Instead of the manual selection used in the common RF method, we adopt the active learning (AL) algorithm to select the samples from the initial results automatically in each RF iteration. Moreover, to ensure the selected images are representative and informative enough, we choose different AL algorithms to complete the different RF processes. Finally, the contributions of all AL-driven RF methods are integrated using a circular fusion scheme. The encouraging experimental results on the ground truth RS image archive illustrate that our CRF is useful for enhancing the performance of RSIR. In addition, compared with many existing RF methods, our CRF achieves improved behavior.