Improve Biomedical Information Retrieval Using Modified Learning to Rank Methods.

Research articles in biomedicine domain have increased exponentially, which makes it more and more difficult for biologists to manually capture all the information they need. Information retrieval technologies can help to obtain the users' needed information automatically. However, it is a great challenge to apply these technologies to biomedicine domain directly because of some domain specific characteristics, such as the abundance of terminologies. To enhance the effectiveness of the biomedical information retrieval, we propose a novel framework based on the state-of-the-art information retrieval methods, called learning to rank, which has been proved effective to rank documents based on their relevance degree. In the framework, we attempt to tackle the problem of the abundance of terminologies by constructing ranking models, which focus on not only retrieving the most relevant documents but also diversifying the searching results to increase the completeness of the resulting list for a given query. In the model training, we propose two novel document labeling strategies, and combine several traditional retrieval models as learning features. Besides, we also investigate the usefulness of different learning to rank approaches in our framework. Experimental results on TREC Genomics datasets demonstrate our proposed framework is effective in improving the performance of biomedical information retrieval.

With the rapid development of biomedicine, the number of biomedical articles has increased accordingly, which presents a great challenge for biologists trying to keep up with the latest research. Information retrieval seeks to meet this challenge by searching among a large number of articles based on given queries and providing the most relevant ones to fulfill information needs. As an effective information retrieval technique, query expansion has some room for improvement to achieve the desired performance when directly applied for biomedical information retrieval because there exist many domain-related terms both in users' queries and in related articles. To solve this problem, we propose a biomedical query expansion framework based on learning-to-rank methods, in which we refine candidate expansion terms by training term-ranking models to select the most relevant terms. To train the term-ranking models, we first propose a pseudo-relevance feedback method based on MeSH to select candidate expansion terms and then represent the candidate terms as feature vectors by defining both the corpus-based term features and the resource-based term features. Experimental results obtained for TREC genomics datasets show that our method can capture more relevant terms to expand the original query and effectively improve biomedical information retrieval performance.

A well-designed information retrieval (IR) framework should provide powerful abstractions for common IR concepts and techniques. Central to the success of such a framework is whether these abstractions can be implemented to satisfy users' demands of functionality, performance, flexibility, and ease of use. This paper discusses implementation issues for a multimedia IR framework in the context of the recently proposed FIRE design. We present the software architecture of the framework, and outline the cornerstone components of its implementation. Issues of reusability, ease of use, and performance are discussed with references to solutions adopted in FIRE.

The method of latent semantic indexing (LSI) is an information retrieval technique using a low-rank singular value decomposition (SVD) of the term-document matrix. Although the LSI method has empirical success, it suffers from the lack of interpretation for the low-rank approximation and, consequently, the lack of controls for accomplishing specific tasks in information retrieval. A method introduced by Dhillon and Modha is an improvement in that direction. It uses centroids of clusters or so called concept decomposition for lowering the rank of the term-document matrix. We focus on improvements of that method using fuzzy k-means algorithm. Also, we compare the precision of information retrieval for the two above methods.

We define a general framework for ontology-based information retrieval (IR). In our approach, document and query expansion rely on a base taxonomy that is extracted from a lexical database or a Linked Data set (e.g. WordNet, Wiktionary etc.). Each term from a document or query is modelled as a vector of base concepts from the base taxonomy. We define a set of mapping functions which map multiple ontological layers (dimensions) onto the base taxonomy. This way, each concept from the included ontologies can also be represented as a vector of base concepts from the base taxonomy. We propose a general weighting schema which is used for the vector space model. Our framework can therefore take into account various lexical and semantic relations between terms and concepts (e.g. synonymy, hierarchy, meronymy, antonymy, geo-proximity, etc.). This allows us to avoid certain vocabulary problems (e.g. synonymy, polysemy) as well as to reduce the vector size in the IR tasks.

The Web has become a large repository of documents (or pages) written in many different languages. In this context, traditional information retrieval (IR) techniques cannot be used whenever the user query and the documents being retrieved are in different languages. To address this problem, new cross‐language information retrieval (CLIR) techniques have been proposed. In this work, we describe a method for cross‐language retrieval of medical information. This method combines query terms and related medical concepts obtained automatically through a categorization procedure. The medical concepts are used to create a linguistic abstraction that allows retrieval of information in a language‐independent way, minimizing linguistic problems such as polysemy. To evaluate our method, we carried out experiments using the OHSUMED test collection, whose documents are written in English, with queries expressed in Portuguese, Spanish, and French. The results indicate that our cross‐language retrieval method is as effective as a standard vector space model algorithm operating on queries and documents in the same language. Further, our results are better than previous results in the literature.

Learning a high-dimensional dense representation for vocabulary terms, also known as a word embedding, has recently attracted much attention in natural language processing and information retrieval tasks. The embedding vectors are typically learned based on term proximity in a large corpus. This means that the objective in well-known word embedding algorithms, e.g., word2vec, is to accurately predict adjacent word(s) for a given word or context. However, this objective is not necessarily equivalent to the goal of many information retrieval (IR) tasks. The primary objective in various IR tasks is to capture relevance instead of term proximity, syntactic, or even semantic similarity. This is the motivation for developing unsupervised relevance-based word embedding models that learn word representations based on query-document relevance information. In this paper, we propose two learning models with different objective functions; one learns a relevance distribution over the vocabulary set for each query, and the other classifies each term as belonging to the relevant or non-relevant class for each query. To train our models, we used over six million unique queries and the top ranked documents retrieved in response to each query, which are assumed to be relevant to the query. We extrinsically evaluate our learned word representation models using two IR tasks: query expansion and query classification. Both query expansion experiments on four TREC collections and query classification experiments on the KDD Cup 2005 dataset suggest that the relevance-based word embedding models significantly outperform state-of-the-art proximity-based embedding models, such as word2vec and GloVe.

Information retrieval (IR) is essential in biomedical knowledge acquisition and clinical decision support. While recent progress has shown that language model encoders perform better semantic retrieval, training such models requires abundant query-article annotations that are difficult to obtain in biomedicine. As a result, most biomedical IR systems only conduct lexical matching. In response, we introduce MedCPT, a first-of-its-kind Contrastively Pre-trained Transformer model for zero-shot semantic IR in biomedicine. To train MedCPT, we collected an unprecedented scale of 255 million user click logs from PubMed. With such data, we use contrastive learning to train a pair of closely integrated retriever and re-ranker. Experimental results show that MedCPT sets new state-of-the-art performance on six biomedical IR tasks, outperforming various baselines including much larger models, such as GPT-3-sized cpt-text-XL. In addition, MedCPT also generates better biomedical article and sentence representations for semantic evaluations. As such, MedCPT can be readily applied to various real-world biomedical IR tasks. The MedCPT code and model are available at https://github.com/ncbi/MedCPT.

Text mining and information retrieval techniques have been developed to assist us with analyzing, organizing and retrieving documents with the help of computers. In many cases, it is desirable that the authors of such documents remain anonymous: Search logs can reveal sensitive details about a user, critical articles or messages about a company or government might have severe or fatal consequences for a critic, and negative feedback in customer surveys might negatively impact business relations if they are identified. Simply removing personally identifying information from a document is, however, insufficient to protect the writer's identity: Given some reference texts of suspect authors, so-called authorship attribution methods can reidentfy the author from the text itself. One of the most prominent models to represent documents in many common text mining and information retrieval tasks is the vector space model where each document is represented as a vector, typically containing its term frequencies or related quantities. We therefore propose an automated text anonymization approach that produces synthetic term frequency vectors for the input documents that can be used in lieu of the original vectors. We evaluate our method on an exemplary text classification task and demonstrate that it only has a low impact on its accuracy. In contrast, we show that our method strongly affects authorship attribution techniques to the level that they become infeasible with a much stronger decline in accuracy. Other than previous authorship obfuscation methods, our approach is the first that fulfills differential privacy and hence comes with a provable plausible deniability guarantee.

Analyzing and modeling user behavior in the Information Retrieval (IR) process is important for IR research. In this paper, we propose a new research schema in IR research. In this schema, we first investigate humans' cognitive process when completing a specific IR task and build cognitive models for that task. The findings in the investigation and the proposed cognitive models are then utilized to improve machines' performance in the IR task. Through this research schema, we revisit three IR tasks. In the first study, we carefully analyze users' clicking behavior in mobile search and propose a mobile click model to extract relevance feedback from the mobile search logs. In the second study, we conduct an eye-tracking study to investigate how human assessors read a document during relevance judgment task and adopt the findings in building a novel retrieval model that can better approximate humans' relevance judgment. In the last study, we conduct another eye-tracking study to investigate humans' reading behavior when completing the reading comprehension task. We build a prediction model for user attention and leverage the predicted attention signals to improve the machine reading comprehension model. By successfully adopting this research schema to three different IR tasks, we demonstrate its effectiveness and generalizability.

Today, the widespread use of mobile devices allows users to search information "on the go»», whenever and wherever they want, no longer confining Information Retrieval to classic desktop interfaces. We believe that technical advances in Augmented Reality will allow Information Retrieval to go even further, making use of both the users» surroundings and their abilities to interact with the physical world. In this paper, we present the fundamental concept of Reality-Based Information Retrieval, which combines the classic Information Retrieval process with Augmented Reality technologies to provide context-dependent search cues and situated visualizations of the query and the results. With information needs often stemming from real-world experiences, this novel combination has the potential to better support both Just-in-time Information Retrieval and serendipity. Based on extensive literature research, we propose a conceptual framework for Reality-Based Information Retrieval. We illustrate and discuss this framework and present two prototypical implementations, which we tested in small user studies. They demonstrate the feasibility of our concepts and inspired our discussion of notable challenges for further research in this novel and promising area.

With the extensive use of computer aided systems in product design, manufacture and analysis, a mass of documents are produced. These documents provide engineering information in integrated environment of enterprise for engineers, but maintained by different systems in different forms. Current information retrieval approaches usually lack semantic supports for different kinds of documents, which leads to insufficiency of content representation and misunderstanding of query intention. To effectively search engineering information in various documents, and consider semantic information in engineering domain, an ontology-based information retrieval framework is presented. As the center of framework, ontologies are established to support document analysis and query processing through information representation in semantic level, and complete the mapping between user queries and document resources. The framework provides a unified platform for multi-source engineering information retrieval (EIR) from various documents in integrated environment.

The Web enables to have access to silo-ed information describing news articles, often offering a multitude of viewpoints that, once combined, can provide a broader picture of the story being reported on the news. In this paper, we propose an approach that automatically extracts representative features of a news item, namely named entities, from textual content attached to a video item (subtitles) and from a set of documents from the Web collected using entity expansion techniques. Approaches relying on entity expansion generally try to collect and process the important facts behinds a particular news item, but they are often too dependent on frequency-based functions and information retrieval techniques thus neglecting the multi-dimensional relationships that are established among the entities. We propose a concentric-based approach that enables to represent the context of a news item, by harmonizing into a single model the representative entities, which can be extracted using information retrieval and natural language processing techniques (Core), and other entities that get prominent according to different dimensions such as informativeness, semantic connectivity, or popularity (Crust). We compare our approach with a baseline by analyzing the compactness of the generated summary on an existing gold standard available on the Web. Results of the experiments show that our approach converges faster to the ideal compact news snapshot with an improvement of 36.9% over the baseline.

This chapter provides a perspective on a problem-based learning (PBL) approach to teaching information retrieval (IR) theory, techniques, and practice at under- and postgraduate levels. The educational philosophy of PBL is outlined, and its pertinence to addressing the challenges of teaching IR in terms of both the diverse nature of the subject and also of the students is discussed. It is argued that teaching IR as a solution to a problem (access to information) and as a problem (how can we organize information in IR systems so it can be accessed) provides an engaging and challenging way to teach it to a range of students. It also draws connections between IR theory and techniques and relates IR to other disciplines through showing how similar problems are faced in other subjects. Some specific examples of teaching exercises based on PBL are provided with a discussion and reflection on their effectiveness. Finally assessment and feedback from students are discussed.

Companies often develop and maintain a collection of product variants that share some common features but also support different, customer-specific features. To reengineering such legacy product variants for systematic reuse, one must identify features and their implementing code units (e.g. functions, files) in different product variants. Information retrieval (IR) techniques may be applied for that purpose. In this paper, we discuss problems that hinder direct application of IR techniques to a collection of product variants. To counter these problems, we present an approach to support effective feature location in product variants. The novelty of our approach is that we exploit commonalities and differences of product variants by software differencing and FCA techniques so that IR technique can achieve satisfactory results for feature location in product variants. We have implemented our approach and conducted evaluation with a collection of nine Linux kernel product variants. Our evaluation shows that our approach always significantly outperforms a direct application of IR technique in the subject product variants.