To improve the performance of the dual-encoder retriever, one effective approach is knowledge distillation from the cross-encoder ranker. Existing works construct the candidate passages following the supervised learning setting where a query is paired with a positive passage and a batch of negatives. However, through empirical observation, we find that even the hard negatives from advanced methods are still too trivial for the teacher to distinguish, preventing the teacher from transferring abundant dark knowledge to the student through its soft label. To alleviate this issue, we propose ADAM, a knowledge distillation framework that can better transfer the dark knowledge held in the teacher with Adaptive Dark exAMples. Different from previous works that only rely on one positive and hard negatives as candidate passages, we create dark examples that all have moderate relevance to the query through mixing-up and masking in discrete space. Furthermore, as the quality of knowledge held in different training instances varies as measured by the teacher's confidence score, we propose a self-paced distillation strategy that adaptively concentrates on a subset of high-quality instances to conduct our dark-example-based knowledge distillation to help the student learn better. We conduct experiments on two widely-used benchmarks and verify the effectiveness of our method.

Recent advances in open-domain question answering (ODQA) have demonstrated impressive accuracy on standard Wikipedia style benchmarks. However, it is less clear how robust these models are and how well they perform when applied to real-world applications in drastically different domains. While there has been some work investigating how well ODQA models perform when tested for out-of-domain (OOD) generalization, these studies have been conducted only under conservative shifts in data distribution and typically focus on a single component (ie. retrieval) rather than an end-to-end system. In response, we propose a more realistic and challenging domain shift evaluation setting and, through extensive experiments, study end-to-end model performance. We find that not only do models fail to generalize, but high retrieval scores often still yield poor answer prediction accuracy. We then categorize different types of shifts and propose techniques that, when presented with a new dataset, predict if intervention methods are likely to be successful. Finally, using insights from this analysis, we propose and evaluate several intervention methods which improve end-to-end answer F1 score by up to 24 points.

Pre-trained language models have achieved promising success in code retrieval tasks, where a natural language documentation query is given to find the most relevant existing code snippet. However, existing models focus only on optimizing the documentation code pairs by embedding them into latent space, without the association of external knowledge. In this paper, we propose a generation-augmented query expansion framework. Inspired by the human retrieval process - sketching an answer before searching, in this work, we utilize the powerful code generation model to benefit the code retrieval task. Specifically, we demonstrate that rather than merely retrieving the target code snippet according to the documentation query, it would be helpful to augment the documentation query with its generation counterpart - generated code snippets from the code generation model. To the best of our knowledge, this is the first attempt that leverages the code generation model to enhance the code retrieval task. We achieve new state-of-the-art results on the CodeSearchNet benchmark and surpass the baselines significantly.

Automated summarization of clinical texts can reduce the burden of medical professionals. "Discharge summaries" are one promising application of the summarization, because they can be generated from daily inpatient records. Our preliminary experiment suggests that 20-31% of the descriptions in discharge summaries overlap with the content of the inpatient records. However, it remains unclear how the summaries should be generated from the unstructured source. To decompose the physician's summarization process, this study aimed to identify the optimal granularity in summarization. We first defined three types of summarization units with different granularities to compare the performance of the discharge summary generation: whole sentences, clinical segments, and clauses. We defined clinical segments in this study, aiming to express the smallest medically meaningful concepts. To obtain the clinical segments, it was necessary to automatically split the texts in the first stage of the pipeline. Accordingly, we compared rule-based methods and a machine learning method, and the latter outperformed the formers with an F1 score of 0.846 in the splitting task. Next, we experimentally measured the accuracy of extractive summarization using the three types of units, based on the ROUGE-1 metric, on a multi-institutional national archive of health records in Japan. The measured accuracies of extractive summarization using whole sentences, clinical segments, and clauses were 31.91, 36.15, and 25.18, respectively. We found that the clinical segments yielded higher accuracy than sentences and clauses. This result indicates that summarization of inpatient records demands finer granularity than sentence-oriented processing. Although we used only Japanese health records, it can be interpreted as follows: physicians extract "concepts of medical significance" from patient records and recombine them ...

In the past few decades, there has been an explosion in the amount of available data produced from various sources with different topics. The availability of this enormous data necessitates us to adopt effective computational tools to explore the data. This leads to an intense growing interest in the research community to develop computational methods focused on processing this text data. A line of study focused on condensing the text so that we are able to get a higher level of understanding in a shorter time. The two important tasks to do this are keyword extraction and text summarization. In keyword extraction, we are interested in finding the key important words from a text. This makes us familiar with the general topic of a text. In text summarization, we are interested in producing a short-length text which includes important information about the document. The TextRank algorithm, an unsupervised learning method that is an extension of the PageRank (algorithm which is the base algorithm of Google search engine for searching pages and ranking them) has shown its efficacy in large-scale text mining, especially for text summarization and keyword extraction. this algorithm can automatically extract the important parts of a text (keywords or sentences) and declare them as the result. However, this algorithm neglects the semantic similarity between the different parts. In this work, we improved the results of the TextRank algorithm by incorporating the semantic similarity between parts of the text. Aside from keyword extraction and text summarization, we develop a topic clustering algorithm based on our framework which can be used individually or as a part of generating the summary to overcome coverage problems.

Medical systematic reviews typically require assessing all the documents retrieved by a search. The reason is two-fold: the task aims for ``total recall''; and documents retrieved using Boolean search are an unordered set, and thus it is unclear how an assessor could examine only a subset. Screening prioritisation is the process of ranking the (unordered) set of retrieved documents, allowing assessors to begin the downstream processes of the systematic review creation earlier, leading to earlier completion of the review, or even avoiding screening documents ranked least relevant. Screening prioritisation requires highly effective ranking methods. Pre-trained language models are state-of-the-art on many IR tasks but have yet to be applied to systematic review screening prioritisation. In this paper, we apply several pre-trained language models to the systematic review document ranking task, both directly and fine-tuned. An empirical analysis compares how effective neural methods compare to traditional methods for this task. We also investigate different types of document representations for neural methods and their impact on ranking performance. Our results show that BERT-based rankers outperform the current state-of-the-art screening prioritisation methods. However, BERT rankers and existing methods can actually be complementary, and thus, further improvements may be achieved if used in conjunction.

Ultra-fine entity typing (UFET) predicts extremely free-formed types (e.g., president, politician) of a given entity mention (e.g., Joe Biden) in context. State-of-the-art (SOTA) methods use the cross-encoder (CE) based architecture. CE concatenates the mention (and its context) with each type and feeds the pairs into a pretrained language model (PLM) to score their relevance. It brings deeper interaction between mention and types to reach better performance but has to perform N (type set size) forward passes to infer types of a single mention. CE is therefore very slow in inference when the type set is large (e.g., N = 10k for UFET). To this end, we propose to perform entity typing in a recall-expand-filter manner. The recall and expand stages prune the large type set and generate K (K is typically less than 256) most relevant type candidates for each mention. At the filter stage, we use a novel model called MCCE to concurrently encode and score these K candidates in only one forward pass to obtain the final type prediction. We investigate different variants of MCCE and extensive experiments show that MCCE under our paradigm reaches SOTA performance on ultra-fine entity typing and is thousands of times faster than the cross-encoder. We also found MCCE is very effective in fine-grained (130 types) and coarse-grained (9 types) entity typing. Our code is available at \url{https://github.com/modelscope/AdaSeq/tree/master/examples/MCCE}.

Recently proposed systems for open-domain question answering (OpenQA) require large amounts of training data to achieve state-of-the-art performance. However, data annotation is known to be time-consuming and therefore expensive to acquire. As a result, the appropriate datasets are available only for a handful of languages (mainly English and Chinese). In this work, we introduce and publicly release PolQA, the first Polish dataset for OpenQA. It consists of 7,000 questions, 87,525 manually labeled evidence passages, and a corpus of over 7,097,322 candidate passages. Each question is classified according to its formulation, type, as well as entity type of the answer. This resource allows us to evaluate the impact of different annotation choices on the performance of the QA system and propose an efficient annotation strategy that increases the passage retrieval performance by 10.55 p.p. while reducing the annotation cost by 82%.

Example: 'I want to read a book' In the above example there are 6 tokens which are- ('I', 'want, 'to', 'read', 'a' and'book') A type is the class of all tokens containing the same character sequence. In the above example, there are only 5 types which are - 'can, 'you', 'a, 'as' and'canner' as'can', 'as' and'a' are being repeated. In the above example, by deleting period and hyphens between the characters and words we are normalizing the type by making it a term. So the term in the above example is: 'USA' and'antiinflammatory' Example: "Hello everyone.Welcome to the course." The tokens for the given sentence will be -- ['Hello','everyone', 'Welcome', 'to', 'the', 'course'] Welcome to the Natural Language Processing course.

As modern data pipelines continue to collect, produce, and store a variety of data formats, extracting and combining value from traditional and context-rich sources such as strings, text, video, audio, and logs becomes a manual process where such formats are unsuitable for RDBMS. To tap into the dark data, domain experts analyze and extract insights and integrate them into the data repositories. This process can involve out-of-DBMS, ad-hoc analysis, and processing resulting in ETL, engineering effort, and suboptimal performance. While AI systems based on ML models can automate the analysis process, they often further generate context-rich answers. Using multiple sources of truth, for either training the models or in the form of knowledge bases, further exacerbates the problem of consolidating the data of interest. We envision an analytical engine co-optimized with components that enable context-rich analysis. Firstly, as the data from different sources or resulting from model answering cannot be cleaned ahead of time, we propose using online data integration via model-assisted similarity operations. Secondly, we aim for a holistic pipeline cost- and rule-based optimization across relational and model-based operators. Thirdly, with increasingly heterogeneous hardware and equally heterogeneous workloads ranging from traditional relational analytics to generative model inference, we envision a system that just-in-time adapts to the complex analytical query requirements. To solve increasingly complex analytical problems, ML offers attractive solutions that must be combined with traditional analytical processing and benefit from decades of database community research to achieve scalability and performance effortless for the end user.