Objective:Patient metadata exist in published articles, but are often disconnected from genome sequences in databases, limiting their utility for genomic epidemiology. The objective of this study was to develop and evaluate natural language processing methods to facilitate the large-scale detection of patient metadata associated with reports of genome sequencing in published articles, drawing on the case of SARS-CoV-2. Methods:We applied filters to select a sample of 245 PubMed articles (50,918 sentences) in LitCovid for manual annotation of sentences that reported generating SARS-CoV-2 sequences. We trained, deployed, and validated a BERT-based classifier, and selected a sample of 150 predicted articles (22,147 sentences) for manual annotation of sentences that reported patient metadata associated with the sequences. In addition to training BERT-based classifiers, we experimented with a generative AI approach, prompting the Llama-3-70B LLM using zero-shot, role-based, few-shot, chain-of-thought, and reasoning-eliciting prompting. Results:BERT-based models that were pre-trained on corpora in biomedical or, more specifically, COVID-19 domains outperformed those that were pre-trained on corpora in general domains for detecting reports of patient metadata associated with SARS-CoV-2 sequences, achieving the best performance with a classifier based on a BiomedBERT-Large-Abstract model (F1-score = 0.776). While the best performance of our generative AI approach was achieved using role-based, few-shot, and chain-of-thought prompting (F1-score = 0.558), it was nonetheless outperformed by all of our machine learning-based classifiers. Conclusion:Our methods were applied to more than 350,000 published articles and can be used to advance the utility and efficiency of genomic epidemiology for public health responses to virus outbreaks.

都市緑地は,多様な役割があり,最近では脱炭素社会および循環型社会構築への貢献も求められている。本研究では,CO2固定のための植栽基盤へのバイオ炭施用が緑化樹木の生育に及ぼす影響を評価することを目的とした。植栽基盤(黒土およびマサ土)に3種類のバイオ炭(木炭,竹炭およびもみ殻くん炭)を施用し,造園樹木の苗木の生育試験を行った。前報では,4ヶ月程の生育期間における結果を報告したが,本研究では1年以上(2022.9~2023.12)の長期の生育期間を経て得られた成果を報告する。得られた結果から,バイオ炭施用区のほとんどで樹木が生育する傾向が認められた。また,成長量は,土壌やバイオ炭施用割合,樹種によって異なることが明らかとなった。

都市緑地は,多様な役割があり,最近では脱炭素社会および循環型社会構築への貢献も求められてきている。本研究では,緑化樹木の植栽基盤へのバイオ炭施用によるCO2固定効果を評価することを目的とした。植栽基盤(黒土およびマサ土)に数種類のバイオ炭(木炭,竹炭およびもみ殻くん炭)を混合し,造園樹木の苗木の生育試験(9月~12月)を行った。得られた結果より,CO2固定効果と緑化樹木の生育効果の最適化を考察した。いずれのバイオ炭を土壌に施用した場合でも,植物生育の明確な阻害は確認されなかった。また,バイオ炭の施用割合別に,植栽基盤に固定することのできるCO2量は,0.03~0.24 t-CO2/m3と試算された。

Constructing a machine that understands human language is one of the most elusive and long-standing challenges in artificial intelligence. This thesis addresses this challenge through studies of reading comprehension with a focus on understanding entities and their relationships. More specifically, we focus on question answering tasks designed to measure reading comprehension. We focus on entities and relations because they are typically used to represent the semantics of natural language.

In this paper, we explore the utility of sentiment analysis and text classification of voice of the customer (VOC) for improving churn prediction, which is a task to detect customers who are about to quit. Our work is motivated by the observation that the increase of customer satisfaction will reproduce churn and the customer satisfaction can be reflected in some degree by applying NLP techniques on VOC, the unstructured textual information which captures a view of customer's attitude and feedbacks. To the best of our knowledge, this is the first work that introduces text classification of VOC to churn prediction task. Experiments show that adding VOC analysis into a conventional churn prediction model results in a significant increase in predictive performance.

Deep Recurrent Neural Network (RNN) has gained popularity in many sequence classification tasks. Beyond predicting a correct class for each data instance, data scientists also want to understand what differentiating factors in the data have contributed to the classification during the learning process. We present a visual analytics approach to facilitate this task by revealing the RNN attention for all data instances, their temporal positions in the sequences, and the attribution of variables at each value level. We demonstrate with real-world datasets that our approach can help data scientists to understand such dynamics in deep RNNs from the training results, hence guiding their modeling process.

In this study, we develop a computer-aided material design system to represent and extract knowledge related to material design from natural language texts. A machine learning model is trained on a text corpus weakly labeled by minimal annotated relationship data (~100 labeled relationships) to extract knowledge from scientific articles. The knowledge is represented by relationships between scientific concepts, such as {annealing, grain size, strength}. The extracted relationships are represented as a knowledge graph formatted according to design charts, inspired by the process-structure-property-performance (PSPP) reciprocity. The design chart provides an intuitive effect of processes on properties and prospective processes to achieve the certain desired properties. Our system semantically searches the scientific literature and provides knowledge in the form of a design chart, and we hope it contributes more efficient developments of new materials.

Takeshi Onishi, Davy Weissenbacher, Ari Klein, Karen O’Connor, Graciela Gonzalez-Hernandez. Proceedings of the 2018 EMNLP Workshop SMM4H: The 3rd Social Media Mining for Health Applications Workshop & Shared Task. 2018.

The human head pose provides insights on the activities or intentions of a given person. Head pose estimation techniques are thus often employed in intelligent surveillance camera systems for marketing analysis or security monitoring. Nowadays, omnidirectional cameras have become widely used in surveillance systems owing to their unique property of wide-range coverage. However, this property causes significant changes in visual appearance and distortions inside the image, and general approaches using a head image may fail in estimation. In this paper, we thus propose a method for head pose estimation using omnidirectional camera images. The proposed model employs both a head image and full body image for cases in which a face is self-occluded and the head image is thus almost useless. In addition, image attribute data are integrated into the network to learn the relation between the changes in appearance or distortion and locations inside the whole image. Experiments are conducted to compare the accuracy of the presented approach with those of ordinary methods. It is verified that the proposed method improves the accuracy by more than 19% over the baseline method.

Sales Performance Management (SPM) solutions for enterprise-grade businesses generate large volumes of multi-dimensional data, including temporal event sequences and tabular attributes. Among attributes of different data structures, it is difficult to find clear connections between factors and outcomes. Discovering key factors and their influences from multivariate data can provide instructional advice to help sales representatives (SR) maintain healthy relationships with customers and achieve sales goals. This paper describes the FactorLink approach for 1) correlating temporal event sequences, multi-dimensional tabular data with their outcomes, 2) interactively assisting users to find key factors and understand their influences, and 3) exploring potential outcomes by reviewing and comparing the patterns found in the integrated SPM data. We conducted several case studies, and the results demonstrate the effectiveness of our approach.

A significant number of neural architectures for reading comprehension have recently been developed and evaluated on large cloze-style datasets. We present experiments supporting the emergence of “predication structure” in the hidden state vectors of these readers. More specifically, we provide evidence that the hidden state vectors represent atomic formulas \Phi[c] where \Phi is a semantic property (predicate) and c is a constant symbol entity identifier.

Reading comprehension is a question answering task where the answer is to be found in a given passage about entities and events not mentioned in general knowledge sources. A significant number of neural architectures for this task (neural readers) have recently been developed and evaluated on large cloze-style datasets. We present experiments supporting the emergence of “predication structure” in the hidden state vectors of a class of neural readers including the Attentive Reader and Stanford Reader. We posits that the hidden state vectors can be viewed as (a representation of) a concatenation [P, c] of a “predicate vector” P and a “constant symbol vector” c and that the hidden state represents the atomic formula P(c). This predication structure plays a conceptual role in relating “aggregation readers” such as the Attentive Reader and the Stanford Reader to “explicit reference readers” such as the Attention-Sum Reader, the Gated-Attention Reader and the Attention-over-Attention Reader. In an independent contribution, we show that the addition of linguistics features to the input to existing neural readers significantly boosts performance yielding the best results to date on the Who-did-What dataset.

Reading comprehension is a question answering task where the answer is to be found in a given passage about entities and events not mentioned in general knowledge sources. A significant number of neural architectures for this task (neural readers) have recently been developed and evaluated on large cloze-style datasets. We present experiments supporting the emergence of “predication structure” in the hidden state vectors of a class of neural readers including the Attentive Reader and Stanford Reader. We posits that the hidden state vectors can be viewed as (a representation of) a concatenation [P, c] of a “predicate vector” P and a “constant symbol vector” c and that the hidden state represents the atomic formula P(c). This predication structure plays a conceptual role in relating “aggregation readers” such as the Attentive Reader and the Stanford Reader to “explicit reference readers” such as the Attention-Sum Reader, the Gated-Attention Reader and the Attention-over-Attention Reader. In an independent contribution, we show that the addition of linguistics features to the input to existing neural readers significantly boosts performance yielding the best results to date on the Who-did-What dataset.

We have constructed a new "Who-did-What" dataset of over 200,000 fill-in-the-gap (cloze) multiple choice reading comprehension problems constructed from the LDC English Gigaword newswire corpus. The WDW dataset has a variety of novel features. First, in contrast with the CNN and Daily Mail datasets (Hermann et al., 2015) we avoid using article summaries for question formation. Instead, each problem is formed from two independent articles --- an article given as the passage to be read and a separate article on the same events used to form the question. Second, we avoid anonymization --- each choice is a person named entity. Third, the problems have been filtered to remove a fraction that are easily solved by simple baselines, while remaining 84% solvable by humans. We report performance benchmarks of standard systems and propose the WDW dataset as a challenge task for the community.

Customers, who switch to competitors or move out from service providers, become critical concerns for companies to retain customer loyalty. Churn prevention through churn prediction, which is a task to detect costumers who are about to quit, is one of the methods to ensure customer loyalty with service providers. To detect and analyze early churn is a proactive step to ensure that existing customers do not move out or switch to the product of competitors[1]. On the other hand, customers voice opinions and advices about some brands, companies, products or services. Today voice of the customer(VOC), capturing a view of customer’s behaviors, needs, and feedbacks, can be obtained through center calls, emails, questionnaire, web reviews or SNS[2]. Therefore, it is attractive to consider exploiting VOC analysis in churn prediction task. However few research has been conducted in this direction. One exception is an investigative study showing that integrating the information of call center emails resulted in an increase in predictive performance[3]. They used a weighted term-by-email matrix to represent a collection of emails and used Latent Semantic Indexing (LSI) via Singular Value Decomposition (SVD) to reduce the matrix to k dimension in order to overcome disadvantages of large and sparse matrix. However, it is not possible to know what value of k will lead to an optimal solution in different business situations. They made great effort to determine the critical k for their task. Their techniques required specialized pre-processing and dimension reduction steps. In this paper, we present a simple and easy-toimplement approach to improve the performance of churn prediction by incorporating VOC analysis into a conventional churn prediction model. We first identify the sentimental polarity of VOC, classify opinion types of VOC, calculate VOC prediction scores and then generate new features by these three kinds of information and train new models. We evaluate the usefulness of our approach in a series of experiments and demonstrate that VOC analysis provides substantial performance gains in churn prediction task.