User Relevance Feedback Research Articles

Correlating <i>pre-search</i> and <i>in-search</i> context to predict search intent for exploratory search

Modern information systems are expected to respond to a wide variety of information needs from users with diverse goals. The topical dimension ('what' the user is searching for) of these information needs is well studied; however, the intent dimension ('why' the user is searching) has received relatively less attention. Traditionally, the intent is an immediate reason, purpose, or goal that motivates the user search, and captured in search contexts (pre-search, in-search, pro-search). An ideal information system would be able to use. This article proposed a novel intent estimation strategy, based on the intuition that captured intent proactively extracts potential results. Captured pre-search context adapts query term proximities within matched results beside document-terms statistics and pseudo-relevance feedback with user-relevance feedback for in-search. The assessment asserts the superior performance of the proposed strategy over equivalent on trade-offs, e.g., novelty, diversity (coverage, topicality), retrieval (precision, recall, F-measure) and exploitation vs. exploration.

A scalable sub-graph regularization for efficient content based image retrieval with long-term relevance feedback enhancement

The goal of content-based image retrieval (CBIR) is to search relevant images through the analysis of image content. Manifold Ranking (MR) and Efficient Manifold Ranking (EMR) method has been successfully applied to content-based image retrieval due to its ability to discover underlying geometrical structure of dataset given the query data. But given the image database is scalable, the graph in MR and EMR cannot be extended or updated as their graph size is fixed. In this paper, to solve the above problem, we consider to formulate a sub-graph based on fixed anchors, instead of constructing the graph based on the while dataset, where the anchors are selected by utilizing conventional k-means method and the sub-graph weight matrix is defined by the similarity between any pair-wise anchors. Since the number of anchors is much smaller than the original dataset, updating the sub-graph is much easier than the original graph of whole dataset. Motivated by such sub-graph construction, we then develop an efficient graph regularization framework to predict the ranking scores for the whole data along the sub-graph, where the ranking score is first propagated from query image to the partial anchors, then from partial anchors to all anchors via the sub-graph and finally to the whole dataset. It can also utilize user relevance feedbacks to update the sub-graph so that the discriminative information can be involved to enhance the retrieval performance in a long term. Extensive simulations verify the effectiveness of the proposed method.

Predicting Search Intent Based on In-Search Context for Exploratory Search

Modern information systems are expected to assist users with diverse goals, via exploiting the topical dimension (‘what' the user is searching for) of information needs. However, the intent dimension (‘why' the user is searching) has preferred relatively lesser for the same intention. Traditionally, the intent is an ‘immediate reason, purpose, or goal' that motivates the user search, and captured in search contexts (Pre-search, In-search, Pro-Search), an ideal information system would be able to use. This article proposes a novel intent estimation strategy; based on the intuition that captured intent, and proactively extracts likely results. The captured ‘Pre-search' context adapts query term proximities within matched results beside document-term statistics and pseudo-relevance feedback with user-relevance feedback for In-search. The assessment asserts the superior performance of the proposed strategy over the equivalent on tradeoffs, e.g., novelty, diversity (coverage, topicality), retrieval (precision, recall, F-measure) and exploitation vs exploration.

Probabilistic and machine learning-based retrieval approaches for biomedical dataset retrieval.

The bioCADDIE dataset retrieval challenge brought together different approaches to retrieval of biomedical datasets relevant to a user’s query, expressed as a text description of a needed dataset. We describe experiments in applying a data-driven, machine learning-based approach to biomedical dataset retrieval as part of this challenge. We report on a series of experiments carried out to evaluate the performance of both probabilistic and machine learning-driven techniques from information retrieval, as applied to this challenge. Our experiments with probabilistic information retrieval methods, such as query term weight optimization, automatic query expansion and simulated user relevance feedback, demonstrate that automatically boosting the weights of important keywords in a verbose query is more effective than other methods. We also show that although there is a rich space of potential representations and features available in this domain, machine learning-based re-ranking models are not able to improve on probabilistic information retrieval techniques with the currently available training data. The models and algorithms presented in this paper can serve as a viable implementation of a search engine to provide access to biomedical datasets. The retrieval performance is expected to be further improved by using additional training data that is created by expert annotation, or gathered through usage logs, clicks and other processes during natural operation of the system. Database URL: https://github.com/emory-irlab/biocaddie

Discriminative Multi-View Interactive Image Re-Ranking.

Given an unreliable visual patterns and insufficient query information, content-based image retrieval is often suboptimal and requires image re-ranking using auxiliary information. In this paper, we propose a discriminative multi-view interactive image re-ranking (DMINTIR), which integrates user relevance feedback capturing users' intentions and multiple features that sufficiently describe the images. In DMINTIR, heterogeneous property features are incorporated in the multi-view learning scheme to exploit their complementarities. In addition, a discriminatively learned weight vector is obtained to reassign updated scores and target images for re-ranking. Compared with other multi-view learning techniques, our scheme not only generates a compact representation in the latent space from the redundant multi-view features but also maximally preserves the discriminative information in feature encoding by the large-margin principle. Furthermore, the generalization error bound of the proposed algorithm is theoretically analyzed and shown to be improved by the interactions between the latent space and discriminant function learning. Experimental results on two benchmark data sets demonstrate that our approach boosts baseline retrieval quality and is competitive with the other state-of-the-art re-ranking strategies.

Content-Based Discovery for Web Map Service using Support Vector Machine and User Relevance Feedback.

Many discovery methods for geographic information services have been proposed. There are approaches for finding and matching geographic information services, methods for constructing geographic information service classification schemes, and automatic geographic information discovery. Overall, the efficiency of the geographic information discovery keeps improving., There are however, still two problems in Web Map Service (WMS) discovery that must be solved. Mismatches between the graphic contents of a WMS and the semantic descriptions in the metadata make discovery difficult for human users. End-users and computers comprehend WMSs differently creating semantic gaps in human-computer interactions. To address these problems, we propose an improved query process for WMSs based on the graphic contents of WMS layers, combining Support Vector Machine (SVM) and user relevance feedback. Our experiments demonstrate that the proposed method can improve the accuracy and efficiency of WMS discovery.

IR_URFS_VF: image recommendation with user relevance feedback session and visual features in vertical image search

In recent years, online shopping has grown exponentially and huge number of images are available online. Hence, it is necessary to recommend various product images to aid the user in effortless and efficient access to the desired products. In this paper, we present image recommendation framework with user relevance feedback session and visual features (IR_URFS_VF) to extract relevant images based on user inputs. User feedback is retrieved from image search history with clicked and un-clicked images. Image features are computed off-line and later used to find relevance between images. The relevance between images is determined by cosine similarity and are ranked based on clicked frequency and similarity score between images. Experiments results show that IR_URFS_VF outperforms CBIR method by providing more relevant ranked images to the user input query.

Image retrieval using indexed histogram of Void-and-Cluster Block Truncation Coding

This paper presents a simple approach to improve the image retrieval accuracy in the Void-and-Cluster Block Truncation Coding compressed domain. The proposed approach directly derives an image descriptor from the Ordered Dither Block Truncation Coding (ODBTC) data stream without performing the decoding process. The Color Histogram Feature (CHF) is generated from the two ODBTC color quantizer, while the Halftoning Local Derivative Pattern (HLDP) is constructed from the ODBTC bitmap image. The similarity between two images are measured from their CHF and HLDP features. Three schemes are involved to improve the image retrieval accuracy, including the similarity weight optimization, feature reweighting, and user relevance feedback optimization. An evolutionary stochastic algorithm is exploited to optimize the similarity weight and feature weight in the nearest neighbor distance computation, as well as in the query update of relevance feedback optimization. Section 5 shows that the proposed scheme yields a promising result, and thus it can be a very effective candidate in addressing the content-based image retrieval and image classification task.

AMORE: design and implementation of a commercial-strength parallel hybrid movie recommendation engine

AMORE is a hybrid recommendation system that provides movie recommendation functionality to video-on-demand subscribers of a major triple-play service provider in Greece. Without any user relevance feedback for movies available, all recommendations are solely based on the users' viewing history. To overcome such limitations as well as the extra problem of user histories that are usually the merger of the preferences of all persons in each household, we have performed extensive experiments with open-source recommendation software such as Apache Mahout and Lens-Kit, as well as with our own implementations of several user-based, item-based, and content-based recommendation algorithms. Our results indicate that our own custom multi-threaded implementation of collaborative filtering combined with a custom content-based algorithm outperforms current state-of-the-art implementations of similar algorithms both in solution quality and in response time by margins exceeding 100 % in terms of recall quality and 6300 % in terms of running time. The hybrid nature of the ensemble allows the system to perform well and to overcome inherent limitations of collaborative filtering, such as various cold-start problems. AMORE has been deployed in a production environment where it has contributed to an increase in the provider's rental profits, while at the same time offers customer retention support.

On cold start for associative tag recommendation

Tag recommendation strategies that exploit term co‐occurrence patterns with tags previously assigned to the target object have consistently produced state‐of‐the‐art results. However, such techniques work only for objects with previously assigned tags. Here we focus on tag recommendation for objects with no tags, a variation of the well‐known \textit{cold start} problem. We start by evaluating state‐of‐the‐art co‐occurrence based methods in cold start. Our results show that the effectiveness of these methods suffers in this situation. Moreover, we show that employing various automatic filtering strategies to generate an initial tag set that enables the use of co‐occurrence patterns produces only marginal improvements. We then propose a new approach that exploits both positive and negative user feedback to iteratively select input tags along with a genetic programming strategy to learn the recommendation function. Our experimental results indicate that extending the methods to include user relevance feedback leads to gains in precision of up to 58% over the best baseline in cold start scenarios and gains of up to 43% over the best baseline in objects that contain some initial tags (i.e., no cold start). We also show that our best relevance‐feedback‐driven strategy performs well even in scenarios that lack user cooperation (i.e., users may refuse to provide feedback) and user reliability (i.e., users may provide the wrong feedback).

ENGLISH

Nowadays data is available in a single click that is from World Wide Web which is a huge repository of resources. Web search engines have a key role in the discovery of relevant information, but this kind of search is usually performed using keywords and the results into context less results. Result presented to user may contain context irrelevant data. This work is one step towards better understanding of user context from available domain specific resources. The contextualized strategy for information retrieval (IR) can be built around user profile, query expansion and relevance feedback. In this proposed system, new concept called terminology which is good representative of particular domain (or subject) is defined and used to classify resources as relevant or irrelevant (non- relevant). This work focuses on improving results of context sensitive web search engine based on shared resources. Query expansion using shared documents is applied to implement contextual search. These resources are ranked based on query submitted by user which brings most relevant document at the top in hierarchy. From top ranked documents, terms has been weighted to identify most related terms for query expansion. In addition the results of the query engine with and without the contextual information will be evaluated automatically without any interface of user.

Interactive Multiscale Classification of High-Resolution Remote Sensing Images

The use of remote sensing images (RSIs) as a source of information in agribusiness applications is very common. In those applications, it is fundamental to identify and understand trends and patterns in space occupation. However, the identification and recognition of crop regions in remote sensing images are not trivial tasks yet. In high-resolution image analysis and recognition, many of the problems are related to the representation scale of the data, and to both the size and the representativeness of the training set. In this paper, we propose a method for interactive classification of remote sensing images considering multiscale segmentation. Our aim is to improve the selection of training samples using the features from the most appropriate scales of representation. We use a boosting-based active learning strategy to select regions at various scales for user's relevance feedback. The idea is to select the regions that are closer to the border that separates both target classes: relevant and non-relevant regions. Experimental results showed that the combination of scales produces better results than isolated scales in a relevance feedback process. Furthermore, the interactive method achieved good results with few user interactions. The proposed method needs only a small portion of the training set to build classifiers that are as strong as the ones generated by a supervised method that uses the whole training set.

Your Relevance Feedback Is Essential: Enhancing the Learning to Rank Using the Virtual Feature Based Logistic Regression

Information retrieval applications have to publish their output in the form of ranked lists. Such a requirement motivates researchers to develop methods that can automatically learn effective ranking models. Many existing methods usually perform analysis on multidimensional features of query-document pairs directly and don't take users' interactive feedback information into account. They thus incur the high computation overhead and low retrieval performance due to an indefinite query expression. In this paper, we propose a Virtual Feature based Logistic Regression (VFLR) ranking method that conducts the logistic regression on a set of essential but independent variables, called virtual features (VF). They are extracted via the principal component analysis (PCA) method with the user's relevance feedback. We then predict the ranking score of each queried document to produce a ranked list. We systematically evaluate our method using the LETOR 4.0 benchmark datasets. The experimental results demonstrate that the proposal outperforms the state-of-the-art methods in terms of the Mean Average Precision (MAP), the Precision at position k (P@k), and the Normalized Discounted Cumulative Gain at position k (NDCG@k).

Integrating Recognition and Retrieval With Relevance Feedback for Spoken Term Detection

Recognition and retrieval are typically viewed as two cascaded independent modules for spoken term detection (STD). Retrieval techniques are assumed to be applied on top of automatic speech recognition (ASR) output, with performance depending on ASR accuracy. We propose a framework that integrates recognition and retrieval and consider them jointly in order to yield better STD performance. This can be achieved either by adjusting the acoustic model parameters (model-based) or by considering detected examples (example-based) using relevance information provided by the user (user relevance feedback) or inferred by the system (pseudo-relevance feedback), either for a given query (short-term context) or by taking into account many previous queries (long-term context). Such relevance feedback approaches have long been used in text information retrieval, but are rarely considered and cannot be directly applied to the retrieval of spoken content. The proposed relevance feedback approaches are specific to spoken content retrieval and are hence very different from those developed for text retrieval, which are applied only to text symbols. We present not only these relevance feedback scenarios and approaches for STD, but also propose a framework to integrate them all together. Preliminary experiments showed significant improvements in each case.

Multimodal retrieval with relevance feedback based on genetic programming

This paper presents a framework for multimodal retrieval with relevance feedback based on genetic programming. In this supervised learning-to-rank framework, genetic programming is used for the discovery of effective combination functions of (multimodal) similarity measures using the information obtained throughout the user relevance feedback iterations. With these new functions, several similarity measures, including those extracted from different modalities (e.g., text, and content), are combined into one single measure that properly encodes the user preferences. This framework was instantiated for multimodal image retrieval using visual and textual features and was validated using two image collections, one from the Washington University and another from the ImageCLEF Photographic Retrieval Task. For this image retrieval instance several multimodal relevance feedback techniques were implemented and evaluated. The proposed approach has produced statistically significant better results for multimodal retrieval over single modality approaches and superior effectiveness when compared to the best submissions of the ImageCLEF Photographic Retrieval Task 2008.

Adaptive information retrieval system via modelling user behaviour

There has been an exponential growth in the volume and variety of information available on the Internet, similarly there has been a significant demand from users’ for accurate information that matches their interests, however, the two are often incompatible because of the effectiveness of retrieving the exact information the user requires. This paper addresses this problem with an adaptive agent-based modelling approach that relies on evolutionary user-modelling. The proposed information retrieval system learns user needs from user-provided relevance feedback. It is proposed that retrieval effectiveness can be improved by applying computational intelligence techniques for modelling information needs, through interactive reinforcement learning. The method combines qualitative (subjective) user relevance feedback with quantitative (algorithmic) measures of the relevance of retrieved documents. An adaptive information retrieval system is developed whose retrieval effectiveness is evaluated using traditional precision and recall.

Adaptive learning for relevance feedback: application to digital mammography.

With the rapid growing volume of images in medical databases, development of efficient image retrieval systems to retrieve relevant or similar images to a query image has become an active research area. Despite many efforts to improve the performance of techniques for accurate image retrieval, its success in biomedicine thus far has been quite limited. This article presents an adaptive content-based image retrieval (CBIR) system for improving the performance of image retrieval in mammographic databases. In this work, the authors propose a new relevance feedback approach based on incremental learning with support vector machine (SVM) regression. Also, the authors present a new local perturbation method to further improve the performance of the proposed relevance feedback system. The approaches enable efficient online learning by adapting the current trained model to changes prompted by the user's relevance feedback, avoiding the burden of retraining the CBIR system. To demonstrate the proposed image retrieval system, the authors used two mammogram data sets: A set of 76 mammograms scored based on geometrical similarity and a larger set of 200 mammograms scored by expert radiologists based on pathological findings. The experimental results show that the proposed relevance feedback strategy improves the retrieval precision for both data sets while achieving high efficiency compared to offline SVM. For the data set of 200 mammograms, the authors obtained an average precision of 0.48 and an area under the precision-recall curve of 0.79. In addition, using the same database, the authors achieved a high pathology matching rate greater than 80% between the query and the top retrieved images after relevance feedback. Using mammographic databases, the results demonstrate that the proposed approach is more accurate than the model without using relevance feedback not only in image retrieval but also in pathology matching while maintaining its effectiveness for online relevance feedback applications.

SU‐GG‐I‐153: A Novel Relevance Feedback Approach for Efficient Mammogram Image Retrieval

Purpose: With the rapid increase of medical databases and PACS (picture archiving and communication systems) image volumes, it became critical to develop new tools to retrieve images from these archives more effectively and efficiently for routine clinical practice. A key factor for successful image retrieval systems lies in the development of efficient methods that can derive high‐level semantic concepts from low‐level image content features. Despite many efforts, an efficient retrieval system for medical images still remains a challenging problem due to their complex contents in relation to disease conditions. Towards this goal we have developed a machine learning framework that utilizes expert's knowledge for online retrieval of relevant mammogram images based on clustered microcalcifications. Method and Materials: We tested a database of 200 clinical mammograms that were taken from 104 patients with known pathology. To automatically retrieve mammograms from the database that are perceptually similar to a given query image, we proposed an efficient relevance feedback algorithm based on support vector machine (SVM) regression and incremental learning that dramatically reduces the processing time, avoiding the retraining of the retrieval system. Results: We conducted various experiments to evaluate the developed algorithm under varying user's interactions. It was observed that our proposed method showed significant superiority over off‐line learning. In particular, in a matching fraction test where the top‐k retrieved images were compared with the disease condition of the query image (benign or malignant) after the user's relevance feedback, the proposed method achieved a matching fraction as high as 82.4%. Conclusions: We have developed a novel relevance feedback method for effective retrieval of relevant mammogram images. Our experimental results demonstrated that the proposed method is not only better in performance, but more computationally efficient than offline SVM based on re‐training, while maintaining its effectiveness for online clinical application.

Adaptive relevance feedback model selection for content-based image retrieval

Owing to the rapid development in computer and network technologies, the volumes of modern image repositories have been overwhelming. In this context, traditional image retrieval based on textual indexing is laborious, thus inviting the implementation of content-based image retrieval (CBIR). Relevance feedback (RF) is an iterative procedure which refines the content-based retrievals utilizing the user's RF marked on retrieved results. Recent research has focused on RF model space optimisation. In this paper, we propose an adaptive RF model selection framework which automatically chooses the best RF model with proper parameter values for the given query. The proposed method combines the visual space and model space approaches in order to simultaneously perform two learning tasks, namely, the query optimisation and model optimisation. The particle swarm optimisation (PSO) paradigm is applied to assist the learning tasks. Experimental results tested on a real-world image database reveal that the proposed method outperforms several existing RF approaches using different techniques. The convergence behaviour of the proposed method is empirically analysed.

Batch Mode Active Learning with Applications to Text Categorization and Image Retrieval

Most machine learning tasks in data classification and information retrieval require manually labeled data examples in the training stage. The goal of active learning is to select the most informative examples for manual labeling in these learning tasks. Most of the previous studies in active learning have focused on selecting a single unlabeled example in each iteration. This could be inefficient, since the classification model has to be retrained for every acquired labeled example. It is also inappropriate for the setup of information retrieval tasks where the user's relevance feedback is often provided for the top K retrieved items. In this paper, we present a framework for batch mode active learning, which selects a number of informative examples for manual labeling in each iteration. The key feature of batch mode active learning is to reduce the redundancy among the selected examples such that each example provides unique information for model updating. To this end, we employ the Fisher information matrix as the measurement of model uncertainty, and choose the set of unlabeled examples that can efficiently reduce the Fisher information of the classification model. We apply our batch mode active learning framework to both text categorization and image retrieval. Promising results show that our algorithms are significantly more effective than the active learning approaches that select unlabeled examples based only on their informativeness for the classification model.