Content uploaded by Armita Zarnegar
Author content
All content in this area was uploaded by Armita Zarnegar on Feb 07, 2022
Content may be subject to copyright.
A preview of the PDF is not available
Answer Passage Ranking Enhancement Using Shallow Linguistic Features
Abstract
Question Answering (QA) systems play an important role in decision support systems. Deep neural network-based passage rankers have recently been developed to more effectively rank likely answer-containing passages for QA purposes. These rankers utilize distributed word or sentence embeddings. Such distributed representations mostly carry semantic relatedness of text units in which explicit linguistic features are under-represented. In this paper, we take novel approaches to combine linguistic features (such as different part-of-speech measures) with distributed sentence representations of questions and passages. The QUASAR-T fact-seeking questions and short text passages were used in our experiments to show that while ensembling of deep relevance measures based on pure sentence embedding with linguistic features using several machine learning techniques fails to improve upon the passage ranking performance of our baseline neural network ranker, the concatenation of the same features within the network structure significantly improves the overall performance of passage ranking for QA.
... In such a pipelined architecture, the effectiveness of passage retrieval and ranking has been shown to consistently have a positive correlation with the overall performance of a QA system [11][12][13]. Recent advances in the QA and passage retrieval domains provide evidence that linguistic features of passages and questions play an important role in improving answer passage retrieval effectiveness [14]. To illustrate this, Table 1 summarizes two examples that show the effect of features such as the number of pronouns, the number of nouns, and the query term coverage, in the correct identification of answers to the respective natural language questions. ...
... every single question. Contrary to existing works in the domain, e.g., those in [11,14], this dynamic passage ranking process, therefore, goes beyond a generic, static model that would be applied across several different questions with the same set of feature importance measures. DEA was originally developed to measure the technical and operational efficiency of Decision Making Units (DMUs) that use multiple inputs to produce multiple outputs [15]. ...
... Such embeddings capture contextual information around any given unit of text in such a way that relationships between semantically related units can be established (for example between words city and Melbourne, laptop and computer). The studies in [11,14,[42][43][44][45][46][47][48], those in [49,50] using Bidirectional Encoder Representations from Transformers (BERT) [51], and the work in [52] using phrase-level learning models and dense representations [53,54] are examples of the utilization of deep neural structures to improve passage retrieval. BERT and dense representations have been used in [55] to overcome the low passage ranking performance issues in the presence of typos in query terms. ...
Question Answering (QA) systems play an important role in today’s human–computer interaction systems. QA performance can be significantly improved using effective answer passage retrieval and ranking techniques. Our focus in this paper is on both non machine learning-based and deep learning-based passage retrieval and ranking systems for QA to leverage linguistic features within the text of questions and passages and improve passage ranking effectiveness. We propose a decoupled linguistic and linear programming-based approach for passage ranking using the Data Envelopment Analysis (DEA) technique to improve over well-established answer passage retrieval techniques. Our method scores passages using information retrieval and deep learning relevance metrics, represents retrieved passages using their relevance scores and several linguistic features, and finally makes use of DEA to re-rank the retrieved list of passages. The high effectiveness and significance of our proposed passage ranking method is demonstrated based on several experiments that we have conducted on a standard benchmark data set.
... Re-ranking with a PLM can be performed with a cross-encoder architecture where the question and passage are passed to an encoder trained to identify their relevance. However, the question-passage pair contains semantic features that can help produce a better re-ranking, see e.g., [16,17]. In this paper, we analyze the use of additional information extraction, from both question and passage and its effect on re-ranking. ...
... A significant difference in the distribution of linguistic features between relevant and irrelevant passages is identified. Linguistic features, like parts of speech, are combined with sentence embedding models using the feature concatenation technique in [17]. This enhances the vector representations of the questions and answer passages. ...
Passage re-ranking in question answering (QA) systems is a method to reorder a set of retrieved passages, related to a given question so that answer-containing passages are ranked higher than non-answer-containing passages. With recent advances in language models, passage ranking has become more effective due to improved natural language understanding of the relationship between questions and answer passages. With neural network models, question-passage pairs are used to train a cross-encoder that predicts the semantic relevance score of the pairs and is subsequently used to rank retrieved passages. This paper reports on the use of open information extraction (OpenIE) triples in the form \({<subject, verb, object>}\) for questions and passages to enhance answer passage ranking in neural network models. Coverage and overlap scores of question-passage triples are studied and a novel loss function is developed using the proposed triple-based features to better learn a cross-encoder model to rerank passages. Experiments on three benchmark datasets are compared to the baseline BERT and ERNIE models using the proposed loss function demonstrating improved passage re-ranking performance.KeywordsPassage re-rankingInformation extractionPassage retrievalLinked open data
The fuzzy sets and Intuitionistic fuzzy sets are very useful concepts to elaborate the vagueness in real world problems. The objective of our study is to apply fuzzy set theory and Intuitionistic fuzzy set theory in decision making process. In this paper, we identify in which society a person has to purchase a house in order to fulfil his requirement to maximum extent. In our study we use intuitionistic fuzzy sets to find a relation between the societies and the parameters. And then we find a relation between a person and the parameters. We calculate Normalized Euclidean distance between two Intuitionistic fuzzy sets to make a decision of purchasing house in a society.
While deep learning (DL) technologies are now pervasive in state-of-the-art Computer Vision (CV) and Natural Language Processing (NLP) applications, only in recent years have these technologies started to sufficiently mature in applications related to wireless communications, a field loosely termed Radio Frequency Machine Learning (RFML). In particular, recent research has shown DL to be an enabling technology for Cognitive Radio (CR) applications as well as a useful tool for supplementing expertly defined algorithms for spectrum awareness applications such as signal detection, estimation, and classification. A major driver for the usage of RFML is that little, to no, a priori knowledge of the intended spectral environment is required, given that there is an abundance of representative raw Radio Frequency (RF) data to facilitate training and evaluation. However, in addition to this fundamental need for sufficient data, there are other key considerations, such as trust, security, and hardware requirements, that must be taken into account before deploying RFML systems in real-world wireless communication applications that largely go unaddressed in the current literature. This paper examines the prior works related to these major research considerations, with focus on the dependencies between them and factors unique to the RFML space.
Fuzzy soft sets represent a generalization of fuzzy sets with considerable application potential in decision-making problems, optimization, and computer science. In the paper, we use the fact that the fuzzy soft sets powerset theory is defined by the monad and, using this theory, we introduce the concept of fuzzy soft relations defined by the monad. Using that general method, we can also introduce the fuzzy soft approximation of fuzzy soft sets defined by fuzzy soft relations. We show that fuzzy soft sets and fuzzy soft approximations can be naturally used in selective color segmentation problem, where reliable and fully automated methods do not yet exist.
This paper proposes three modifications for the maximizing model of spherical Bezdek-type fuzzy c -means clustering (msbFCM). First, we use multi-medoids instead of centroids (msbFMMdd), which is similar to modifying fuzzy c -means to fuzzy multi-medoids. Second, we kernelize msbFMMdd (K-msbFMMdd). msbFMMdd can only be applied to objects in the first quadrant of the unit hypersphere, whereas its kernelized form can be applied to a wider class of objects. The third modification is a spectral clustering approach to K-msbFMMdd using a certain assumption. This approach improves the local convergence problem in the original algorithm. Numerical examples demonstrate that the proposed methods can produce good results for clusters with nonlinear borders when an adequate parameter value is selected.
The motivation of the work in this paper is due to the need in research and applied fields for synthetic social network data due to (i) difficulties to obtain real data and (ii) data privacy issues of the real data. The issues to address are first to obtain a graph with a social network type structure, label it with communities. The main focus is the generation of realistic data, its assignment to and propagation within the graph. The main aim in this work is to implement an easy to use standalone end-user application which addresses the aforementioned issues. The methods used are the R-MAT and Louvain algorithms, with some modifications, for graph generation and community labeling respectively, and the development of a Java based system for the data generation using an original seed assignment algorithm followed by a second algorithm for weighted and probabilistic data propagation to neighbors and other nodes. The results show that a close fit can be achieved between the initial user specification and the generated data, and that the algorithms have potential for scale up. The system is made publicly available in a Github Java project.
Scarcity of annotated images hampers the building of automated solution for reliable COVID-19 diagnosis and evaluation from CT. To alleviate the burden of data annotation, we herein present a label-free approach for segmenting COVID-19 lesions in CT via voxel-level anomaly modeling that mines out the relevant knowledge from normal CT lung scans. Our modeling is inspired by the observation that the parts of tracheae and vessels, which lay in the high-intensity range where lesions belong to, exhibit strong patterns. To facilitate the learning of such patterns at a voxel level, we synthesize ‘lesions’ using a set of simple operations and insert the synthesized ‘lesions’ into normal CT lung scans to form training pairs, from which we learn a normalcy-recognizing network (NormNet) that recognizes normal tissues and separate them from possible COVID-19 lesions. Our experiments on three different public datasets validate the effectiveness of NormNet, which conspicuously outperforms a variety of unsupervised anomaly detection (UAD) methods.