Current Information Retrieval Research Articles

Towards Robust Neural Rankers with Large Language Model: A Contrastive Training Approach

Pre-trained language model-based neural rankers have been widely applied in information retrieval (IR). However, the robustness issue of current IR models has not received sufficient attention, which could significantly impact the user experience in practical applications. In this study, we focus on the defensive ability of IR models against query attacks while guaranteeing their retrieval performance. We discover that improving the robustness of IR models not only requires a focus on model architecture and training methods but is also closely related to the quality of data. Different from previous research, we use large language models (LLMs) to generate query variations with the same intent, which exhibit richer and more realistic expressions while maintaining consistent query intent. Based on LLM-generated query variations, we propose a novel contrastive training framework that substantially enhances the robustness of IR models to query perturbations. Specifically, we combine the contrastive loss in the representation space of query variations with the ranking loss in the retrieval training stage to improve the model’s ability to understand the underlying semantic information of queries. Experimental results on two public datasets, WikiQA and ANTIQUE, demonstrate that the proposed contrastive training approach effectively improves the robustness of models facing query attack scenarios while outperforming baselines in retrieval performance. Compared with the best baseline approach, the improvements in average robustness performance of Reranker IR models are 24.9%, 26.5%, 27.0%, and 75.0% on WikiQA and 8.7%, 1.9%, 6.3%, and 13.6% on ANTIQUE, in terms of the MAP (Mean Average Precision), MRR (Mean Reciprocal Rank), nDCG@10 (Normalized Discounted Cumulative Gain) and P@10 (Precision), respectively.

Research Progress and Hotspot of Information Retrieval Correlation based on CiteSpace

Taking 1760 journal papers of information retrieval relevance in the core database of Web of Science as the research object, using literature co-citation network and keyword co-occurrence network analysis, with information visualization as the means, this paper summarizes the research on the relevance of information retrieval. It is found that the current information retrieval correlation research network is concentrated, which is mainly divided into two main knowledge groups, retrieval algorithm and correlation cognition, with few frontier branches. And the knowledge fusion between the two groups needs to be strengthened. Since the emergence of natural language processing, information retrieval relevance has been improved along the path of natural language processing-machine learning-deep learning.

A Hybrid Approach to Discover Entity Synonyms

A major part of the current web information retrieval involves the fulfilment of people's daily needs based upon finding entities involving movies, events, persons, places, concepts, etc. The trend has been the outcome of the tremendous amount of precise and detailed information available on the web making it possible to retrieve the precisely targeted information involving specific instance of a particular domain called entity. The major problem in the entity-based search is the diversity of the name references made to the same instance of the same domain by the content creators and the searchers. These diversified references are known as entity synonyms. The problem is that they cannot be handled through lexical resources. This paper takes into account the previous works carried out in this direction and proposes an advanced mechanism that creates a richer set of entity synonyms than the previous approaches. The work also proposes an index to assess the quality of entity synonyms so generated which is further normalized and fuzzified for implementing automated search.

Batch Evaluation Metrics in Information Retrieval: Measures, Scales, and Meaning

A sequence of recent papers, including in this journal, has considered the role of measurement scales in information retrieval (IR) experimentation, and presented the argument that (only) uniform-step interval scales should be used. Hence, it has been argued, well-known metrics such as reciprocal rank, expected reciprocal rank, normalized discounted cumulative gain, and average precision, should be either discarded as measurement tools, or adapted so that their metric values lie at uniformly-spaced points on the number line. These papers paint a rather bleak picture of past decades of IR evaluation, at odds with the IR community’s overall emphasis on practical experimentation and measurable improvement. Our purpose in this work is to challenge that pessimistic assessment. In particular, we argue that mappings from categorical and ordinal data to sets of points on the number line are valid provided there is an external reason for each target point to have been selected. We first consider the general role of measurement scales, and of categorical, ordinal, interval, ratio, and absolute data collections. In connection with the first two of those categories we also provide examples of the knowledge that is captured and represented by numeric mappings to the real number line. Focusing then on information retrieval, we argue that document rankings are categorical data, and that the role of an effectiveness metric is to provide a single value that summarizes the usefulness to a user or population of users of any given ranking, with usefulness able to be represented as a continuous variable on a ratio scale. That is, we argue that most current IR metrics are well-founded, and, moreover, that those metrics are more meaningful in their current form than in the proposed “intervalized” versions.

A User-Centric Multi-Context Hybrid Reasoning Information Retrieval Model

Information Retrieval has been in existence since the 1940s and is impossible to do without. However, current information retrieval systems are known to be ineffective. The reason that seems to encompass this ineffectiveness is design-focused. Designers and developers of Information Retrieval Systems are known to be system focused rather than user-focused. They provide the same information to users even when they are in different contexts and have diverse preferences. Using a restaurant use case, we propose a conceptual user-centric multi-context hybrid reasoning Information Retrieval model to improve the accuracy of retrieved results.

Efficient Self-Supervised Metric Information Retrieval: A Bibliography Based Method Applied to COVID Literature.

The literature on coronaviruses counts more than 300,000 publications. Finding relevant papers concerning arbitrary queries is essential to discovery helpful knowledge. Current best information retrieval (IR) use deep learning approaches and need supervised training sets with labeled data, namely to know a priori the queries and their corresponding relevant papers. Creating such labeled datasets is time-expensive and requires prominent experts’ efforts, resources insufficiently available under a pandemic time pressure. We present a new self-supervised solution, called SUBLIMER, that does not require labels to learn to search on corpora of scientific papers for most relevant against arbitrary queries. SUBLIMER is a novel efficient IR engine trained on the unsupervised COVID-19 Open Research Dataset (CORD19), using deep metric learning. The core point of our self-supervised approach is that it uses no labels, but exploits the bibliography citations from papers to create a latent space where their spatial proximity is a metric of semantic similarity; for this reason, it can also be applied to other domains of papers corpora. SUBLIMER, despite is self-supervised, outperforms the Precision@5 (P@5) and Bpref of the state-of-the-art competitors on CORD19, which, differently from our approach, require both labeled datasets and a number of trainable parameters that is an order of magnitude higher than our.

Peer review of searches for studies for health technology assessments, systematic reviews, and other evidence syntheses.

Peer review of searches is a process whereby both the search strategies and the search process description are reviewed, ideally using an evidence-based checklist. As the search strategy underpins any well-conducted evidence synthesis, its quality could affect the final result. Evidence shows, however, that search strategies are prone to error. There is increasing awareness and use of the PRESS Evidence-Based Checklist and peer review of search strategies, at the outset of evidence syntheses, prior to the searches being run, and this is now recommended by a number of evidence synthesis organizations. Searches for evidence syntheses should be peer reviewed by a suitably qualified and experienced librarian or information specialist after being designed, ideally, by another suitably qualified and experienced librarian or information specialist. Peer review of searches should take place at two important stages in the evidence synthesis process; at the outset of the project prior to the searches being run and at the prepublication stage. There is little empirical evidence, however, to support the effectiveness of peer review of searches. Further research is required to assess this. Those wishing to stay up to date with the latest developments in information retrieval, including peer review of searches, should consult the SuRe Info resource (http://www.sure-info.org), which seeks to help information specialists and others by providing easy access to the findings from current information retrieval methods research and thus support more research-based information retrieval practice.

Ontology-Based Analysis of Website Structure for Benchmarking in Retail Business

With the growing trend of digital transformation, electronic business has become a crucial part of business operations for many companies. Some of the most critical steps in successful transformation pertain not only to data and information acquisition and digitalization but also to adequate publishing and organization of these information resources. Companies share information through their websites, so the structuring of the web content becomes critical during the digital transformation of business. In this paper, the authors present a conceptual model of the managerial tool that is based on semantic analysis of web sites in order to obtain information about the structure of web sites in a particular domain. The resulting ontological model contains information about best practices in web information organization and can be a valuable resource for management when deciding on the organization of their own web content. The system is based on grounded theory and uses current information retrieval methods, natural language processing, semantic networks, and ontologies.

An Innovative Information Retrieval Model Implementing Particle Swarm Optimization Technique

The Internet has enormous information and it is growing rapidly. The vast amount of data creates challenges in relation to effective Information Retrieval (IR). The scope of the Information Retrieval System (IRS) is to provide the most relevant data for user query from large datasets. However the current IR system fails to provide the hidden and up to date data. This paper focused on soft computing techniques to overcome the above mentioned issues. Particle Swarm Optimization (PSO) is used to compute the fitness function to optimize the retrieval result. PSO has an efficient capability in global search and the implementation is easy to develop. The implementation result of the present study is feasible, that improves the retrieval effect and the accuracy of hidden data retrieval.

Information Retrieval Method for Mixed Intrusion of Surveillance Network Based on Big Data

The current surveillance network intrusion information retrieval methods have problems such as low retrieval accuracy, high recall rate, and long retrieval time. Therefore, this paper proposes a hybrid intrusion information retrieval method for monitoring networks based on big data. The genetic algorithm is used to optimize the feature set, and the partial F test is introduced. The optimal subset is selected to form the optimal feature set and the redundant information elimination model is constructed to eliminate redundant information in the mixed intrusion information. Based on the information retrieval theory, the LDA model is used to build the feature set, and model the subject of the document, establish an intrusion information retrieval model, then complete the hybrid network intrusion information retrieval of the monitoring network under big data. The results show that the method proposed in this paper has higher retrieval accuracy, can effectively improve the retrieval efficiency of intrusion information and the average recall rate is about 24%, which is better than other methods.

Search and analytics using semantic annotations

Search and analytics using semantic annotations

Information retrieval algorithm of industrial cluster based on vector space

AbstractThe current information retrieval research on industrial clusters has low precision, low recall ratio, obvious delay and high energy consumption. Thus, in this paper, a information retrieval algorithm based on vector space for industrial clusters is proposed. By optimizing the unlawful labels in the database network, dividing the web pages of the industrial cluster information database and calculating the keyword scores of the relevant information of the industrial cluster corresponding to a web page, a set of well-divided database pages is obtained, and the purification of the industrial cluster information database is realized. According to the purification of industrial cluster information database, RFD algorithm is used to extract the page data features of purified industrial cluster information database. The extracted results are substituted into the information retrieval, and the vectors composed of retrieval units are used to describe the information of various types of industrial clusters and each retrieval. The matching results of information retrieval are obtained by calculating the correlation between the information of industrial clusters and the query, and the information retrieval of industrial clusters is completed. Experimental results show that the algorithm has high precision and recall ratio, short retrieval time and low energy consumption.

COMPARATIVE ANALYSIS OF RANKING FUNCTIONS FOR RETRIEVING INFORMATION FROM MEDICAL REPOSITORY

Current and forthcoming Information Retrieval algorithms demand high mean average precision with contemporary high recall rates in the technical literature. Nevertheless, the existing state-of-the-art is still not optimized for speed, average query latency, and performance. The previous researchers presented various information retrieval models in the literature but the user search led to a ranking of documents that were hopeful to be relevant. In this paper, an evaluation of various information retrieval models is presented with a range of algorithms. The aim is to elaborate and review the current information retrieval function in the context of enterprise domain- specific search. Experiments were conducted on the OHSUMED benchmark data set from MEDLINE, a medical information database. The experimental results demonstrate that BM25F ranking function outperforms other extensively used ranking functions such as BM25, TFIDF, and Cosine on precision and recall measures.

What Should We Teach in Information Retrieval?

Modern Information Retrieval (IR) systems, such as search engines, recommender systems, and conversational agents, are best thought of as interactive systems. And their development is best thought of as a two-stage development process: offline development followed by continued online adaptation and development based on interactions with users. In this opinion paper, we take a closer look at existing IR textbooks and teaching materials, and examine to which degree they cover the offline and online stages of the IR system development process. We notice that current teaching materials in IR focus mostly on search and on the offline development phase. Other scenarios of interacting with information are largely absent from current IR teaching materials, as is the (interactive) online development phase. We identify a list of scenarios and a list of topics that we believe are essential to any modern set of IR teaching materials that claims to fully cover IR system development. In particular, we argue for more attention, in basic IR teaching materials, to scenarios such as recommender systems, and to topics such as query and interaction mining and understanding, online evaluation, and online learning to rank.

Fairness and Transparency in Ranking

Ranking in Information Retrieval (IR) has been traditionally evaluated from the perspective of the relevance of search engine results to people searching for information, i.e., the extent to which the system provides "the right information, to the right people, in the right way, at the right time." However, people in current IR systems are not only the ones issuing search queries, but increasingly they are also the ones being searched. This raises several new problems in IR that have been addressed in recent research, particularly with respect to fairness/non-discrimination, accountability, and transparency. This is a summary of some these initial developments.

Visual content-based web page categorization with deep transfer learning and metric learning

The growing amounts of online multimedia content challenge the current search, recommendation and information retrieval systems. Information in the form of visual elements is highly valuable in a range of web mining tasks. However, the mining of these resources is a difficult task due to the complexity and variability of images, and the cost of collecting big enough datasets to successfully train accurate deep learning models. This paper proposes a novel framework for the categorization of web pages on the basis of their visual content. This is achieved by exploring the joint application of a transfer learning strategy and metric learning techniques to build a Deep Convolutional Neural Network (DCNN) for feature extraction, even when training data is scarce. The obtained experimental results evidence that the proposed approach outperforms the state-of-the-art handcrafted image descriptors and achieves a high categorization accuracy. In addition, we address the problem of over-time learning, so the proposed framework can learn to identify new web page categories as new labeled images are provided at test time. As a result, prior knowledge of the complete set of possible web categories is not necessary in the initial training phase.

Mixed language queries in online searches

PurposeWith the increasing number of online multilingual resources, cross-language information retrieval (CLIR) has drawn much attention from the information retrieval (IR) research community. However, few studies have examined how and why multilingual searchers seek information in two or more languages, specifically how they switch and mix language in queries to get satisfying results. The purpose of this paper is to focus on Chinese–English bilinguals’ intra-sentential code-switching behaviors in online searches. The scenarios and reasons of code-switching, factors that may affect code-switching, the patterns of mixed language query formulation and reformulation and how current IR systems and other search tools can facilitate such information needs were examined.Design/methodology/approachIn-depth semi-structured interviews were used as the research method. In total, 30 participants were recruited based on their English proficiency, location and profession, using a purposive sampling method.FindingsFour scenarios and four reasons for using Chinese–English mixed language queries to cover information needs were identified, and results suggest that linguistic and cultural/social factors are of equivalent importance in code-switching behaviors. English terms and Chinese terms in queries play different roles in searches, and mixed language queries are irreplaceable by either single language queries or other search facilitating features. Findings also suggest current search engines and tools need greater emphasis in the user interface and more user education is required.Originality/valueThis study presents a qualitative analysis of bilinguals’ code-switching behaviors in online searches. Findings are expected to advance the theoretical understanding of bilingual users’ search strategies and interactions with IR systems, and provide insights for designing more effective IR systems and tools to discover multilingual online resources, including cross-language controlled vocabularies, personalized CLIR tools and mixed language query assistants.

A Survey of Information Retrieval Techniques

The explosive growth of resources stored in various forms and transmitted over the internet has necessitated researches into information retrieval technologies. The major information retrieval mechanisms commonly employed include vector space model, Boolean model, Fuzzy Set model, and probabilistic retrieval model. These models are used to find similarities between the query and the documents to retrieve documents that reflect the query. These approaches are based on key-word, which uses lists of keywords to describe the information content. In this paper, a survey of these models is provided in order to understand their working mechanisms and shortcomings. This understanding is vital as it facilitates the choice of an information retrieval technique, based on the underlying requirements. The results of this survey revealed that the current information retrieval models fall short of the expectations in one way or the other. As such, they are not ideal for high precision information retrieval applications.

Harnessing Music-Related Visual Stereotypes for Music Information Retrieval

Over decades, music labels have shaped easily identifiable genres to improve recognition value and subsequently market sales of new music acts. Referring to print magazines and later to music television as important distribution channels, the visual representation thus played and still plays a significant role in music marketing. Visual stereotypes developed over decades that enable us to quickly identify referenced music only by sight without listening. Despite the richness of music-related visual information provided by music videos and album covers as well as T-shirts, advertisements, and magazines, research towards harnessing this information to advance existing or approach new problems of music retrieval or recommendation is scarce or missing. In this article, we present our research on visual music computing that aims to extract stereotypical music-related visual information from music videos. To provide comprehensive and reproducible results, we present the Music Video Dataset, a thoroughly assembled suite of datasets with dedicated evaluation tasks that are aligned to current Music Information Retrieval tasks. Based on this dataset, we provide evaluations of conventional low-level image processing and affect-related features to provide an overview of the expressiveness of fundamental visual properties such as color, illumination, and contrasts. Further, we introduce a high-level approach based on visual concept detection to facilitate visual stereotypes. This approach decomposes the semantic content of music video frames into concrete concepts such as vehicles, tools, and so on, defined in a wide visual vocabulary. Concepts are detected using convolutional neural networks and their frequency distributions as semantic descriptions for a music video. Evaluations showed that these descriptions show good performance in predicting the music genre of a video and even outperform audio-content descriptors on cross-genre thematic tags. Further, highly significant performance improvements were observed by augmenting audio-based approaches through the introduced visual approach.

Searching bibliographic data using graphs: A visual graph query interface

Abstract With the ever-increasing scientific literature, improving the efficiency of searching bibliographic data has become an important issue. With a lack of support of current bibliographic information retrieval systems in expressing complicated information needs, getting relevant bibliographic data is a demanding task. In this paper, we propose a visual graph query interface for bibliographic information retrieval. Through this interface, users can formulate bibliographic queries by interacting with a graph. Visual graph queries use a set of nodes with constraints and links among nodes to represent explicit and precise bibliographic information needs. The proposed visual graph query interface allows users to formulate several complex bibliographic queries (e.g., bibliographic coupling) that are not attainable in current major bibliographic information retrieval systems. In addition, the proposed interface requires less number of queries in completing everyday bibliographic search tasks.