Search for: All records

Deep neural networks are widely used for text pair classification tasks such as as adhoc retrieval. These deep neural networks are not inherently interpretable and require additional efforts to get rationale behind their decisions. Existing explanation models are not yet capable of inducing alignments between the query terms and the document terms -- which part of the document rationales are responsible for which part of the query? In this paper, we study how the input perturbations can be used to infer or evaluate alignments between the query and document spans, which best explain the black-box ranker’s relevance prediction. We use different perturbation strategies and accordingly propose a set of metrics to evaluate the faithfulness of alignment rationales to the model. Our experiments show that defined metrics based on substitution-based perturbation are more successful in preferring higher-quality alignments, compared to the deletion-based metrics. 

We propose equi-explanation maps to study the variation in model logic across the input space. These global model-agnostic structures partition the hyper-space of explanation features into regions of similar model logic. Equi-explanation maps act as a concise summary of instance explanations and can provide laymen an at-a-glance understanding of the basis on which the classifier makes its decisions. We thus propose the task of generating $\epsilon$-equi-explanation maps, a partitioning of the input space into subspaces such that the standard deviation of explanation vectors in a subspace do not exceed $\epsilon$. We adapt existing local and subspace explainability techniques like LIME and MUSE to generate equi-explanation maps on two binary classification datasets using four classification models and evaluate the quality of their partitioning. We find that these techniques produce a sub-optimal number of subspaces (making the maps harder to interpret) and have a considerable run time. We then propose E-map, a new divide-and-conquer based algorithm to produce $\epsilon$-equi-explanation maps. E-map is able to decrease the number of subspaces (and hence increase interpretability) and running time as compared to the previous systems for a fixed value of $\epsilon$. Finally, given a classifier decision boundary, we try to determine what would be an optimal value for the parameter $\epsilon$. We believe good explanation representation methods can increase the trustworthiness and understanding of machine learning models for critical real world tasks. 

Transformer-based rankers have shown the state-of-the-art performance, but their self-attention operation is mostly unable to process long sequences. One of the common approaches to train these rankers is to heuristically select some segments of each document, such as the first segment, as training data. However, these segments may not contain the query-related parts of documents. To address this problem, we propose the query-driven segment selection from long documents to build training data for transformer-based rankers. The segment selector provides relevant samples with more accurate labels and non-relevant samples which are harder to be predicted. The experimental results show that the basic BERT-based ranker trained with the proposed segment selector significantly outperforms that trained by the heuristically selected segments, and performs equally to the state-of-the-art model with localized self-attention that can process longer input sequences. We also demonstrate that training with our segment selector, there is not much gain from feeding input sequences larger than 200 words. Our findings open up new opportunities to design efficient transformer-based rankers. 

Query Biased Summarization (QBS) aims to produce a summary of the documents retrieved against a query to reduce the human effort for inspecting the full-text content of a document. Typical summarization approaches extract a document text snippet that has term overlap with the query and show that to a searcher. While snippets show relevant information in a document, to the best of our knowledge, there does not exist a summarization system that shows what relevant concepts is missing in a document. Our study focuses on the reduction of user effort in finding relevant documents by exposing them to omitted relevant information. To this end, we use a classical approach, DSPApprox, to find terms or phrases relevant to a query. Then we identify which terms or phrases are missing in a document, present them in a search interface, and ask crowd workers to judge document relevance based on snippets and missing information. Experimental results show both benefits and limitations of this approach. 

Inferring the set name of semantically grouped entities is useful in many tasks related to natural language processing and information retrieval. Previous studies mainly draw names from knowledge bases to ensure high quality, but that limits the candidate scope. We propose an unsupervised framework, AutoName, that exploits large-scale text corpora to name a set of query entities. Specifically, it first extracts hypernym phrases as candidate names from query-related documents via probing a pre-trained language model. A hierarchical density-based clustering is then applied to form potential concepts for these candidate names. Finally, AutoName ranks candidates and picks the top one as the set name based on constituents of the phrase and the semantic similarity of their concepts. We also contribute a new benchmark dataset for this task, consisting of 130 entity sets with name labels. Experimental results show that AutoName generates coherent and meaningful set names and significantly outperforms all compared methods. Further analyses show that AutoName is able to offer explanations for extracted names using the sentences most relevant to the corresponding concept. 

Natural language inference (NLI) is the task of detecting the existence of entailment or contradiction in a given sentence pair. Although NLI techniques could help numerous information retrieval tasks, most solutions for NLI are neural approaches whose lack of interpretability prohibits both straightforward integration and diagnosis for further improvement. We target the task of generating token-level explanations for NLI from a neural model. Many existing approaches for token-level explanation are either computationally costly or require additional annotations for training. In this article, we first introduce a novel method for training an explanation generator that does not require additional human labels. Instead, the explanation generator is trained with the objective of predicting how the model’s classification output will change when parts of the inputs are modified. Second, we propose to build an explanation generator in a multi-task learning setting along with the original NLI task so the explanation generator can utilize the model’s internal behavior. The experiment results suggest that the proposed explanation generator outperforms numerous strong baselines. In addition, our method does not require excessive additional computation at prediction time, which renders it an order of magnitude faster than the best-performing baseline. 

Entity set expansion (ESE) refers to mining ``siblings'' of some user-provided seed entities from unstructured data. It has drawn increasing attention in the IR and NLP communities for its various applications. To the best of our knowledge, there has not been any work towards a supervised neural model for entity set expansion from unstructured data. We suspect that the main reason is the lack of massive annotated entity sets. In order to solve this problem, we propose and implement a toolkit called {DBpedia-Sets}, which automatically extracts entity sets from any plain text collection and can provide a large number of distant supervision data for neural model training. We propose a two-channel neural re-ranking model {NESE} that jointly learns exact and semantic matching of entity contexts. The former accepts entity-context co-occurrence information and the latter learns a non-linear transformer from generally pre-trained embeddings to ESE-task specific embeddings for entities. Experiments on real datasets of different scales from different domains show that {NESE} outperforms state-of-the-art approaches in terms of precision and MAP, where the improvements are statistically significant and are higher when the given corpus is larger. 

Aspect category detection (ACD) is one of the challenging sub-tasks in aspect-based sentiment analysis. The goal of this task is to detect implicit or explicit aspect categories from the sentences of user-generated reviews. Since annotation over the aspects is time-consuming, the amount of labeled data is limited for super-vised learning. In this paper, we study contextual representations of reviews using the BERT model to better extract useful features from text segments in the reviews, and train a supervised classifier with a small amount of labeled data for the ACD task. Experimental results obtained on Amazon reviews of six product domains show that our method is effective in some domains. 

Alerting users that a web page is controversial has been proposed as one method to support critical thinking about text and discourse. We propose an approach to discover controversial topics in a generic document using unsupervised training. Our approach comprises iterative training of a controversy classifier using a disagreement signal within comments and explaining the controversy of the document by generating a topic phrase describing it. Experiments show the effectiveness of our proposed training method using an EM algorithm. When controversial topic extraction is restricted to quality phrases and incorporates TextRank signals, it outperforms several baseline approaches.