Computer-aided education systems are now seeking to provide each student with personalized materials based on a student's individual knowledge. To suitable learning materials, tracing knowledge over period of time is important. However, predicting difficult because students tend forget. The forgetting behavior mainly two reasons: the lag from previous interaction, and number past trials question. Although there few studies that consider while modeling knowledge, some models only partial...

With the rapidly growing use of electronic health records, possibility large-scale clinical information extraction has drawn much attention. We aim to extract adverse drug events and effects from records. As first step this challenge, study assessed (1) how adverse-effect is contained in (2) automatic extracting accuracy current standard Natural Language Processing (NLP) system. Results revealed that 7.7% records include event information, 59% them (4.5% total) can be extracted...

This paper describes the system that has been used by TeamX in SemEval-2014 Task 9 Subtask B. The is a sentiment analyzer based on supervised text categorization approach designed with following two concepts.Firstly, since lexicon features were shown to be effective SemEval-2013 2, various lexicons and pre-processors for them are introduced enhance lexical information.Secondly, distribution of tweets known unbalanced, an weighting scheme bias output machine learner.For test run, was tuned...

We propose a novel geolocation prediction model using complex neural network. Geolocation in social media has attracted many researchers to use information of various types. Our unifies text, metadata, and user network representations with an attention mechanism overcome previous ensemble approaches. In evaluation two open datasets, the proposed exhibited maximum 3.8% increase accuracy 6.6% accuracy@161 against models. further analyzed several intermediate layers our model, which revealed...

Recently, neural models have shown superior performance over conventional in NER tasks. These use CNN to extract sub-word information along with RNN predict a tag for each word. However, these been tested almost entirely on English texts. It remains unclear whether they perform similarly other languages. We worked Japanese using and discovered two obstacles of the state-of-the-art model. First, is unsuitable extracting information. Secondly, model predicting word cannot an entity when part...

Increasing users' positive interactions, such as purchases or clicks, is an important objective of recommender systems. Recommenders typically aim to select items that users will interact with. If the recommended are purchased, increase in sales expected. However, could have been purchased even without recommendation. Thus, we want recommend results caused by This can be formulated a ranking problem terms causal effect. Despite its importance, this has not well explored related research. It...

With the rapidly growing use of electronic health records, possibility large-scale clinical information extraction has drawn much attention. It is not, however, easy to extract because these reports are written in natural language. To address this problem, paper presents a system that converts medical text into table structure. This system's core technologies (1) event recognition modules and (2) negative identification module judges whether an actually occurred or not. Regarding latter...

The amount of medical and clinical-related information on the Web is increasing. Among different types available, social media-based data obtained directly from people are particularly valuable attracting significant attention. To encourage natural language processing (NLP) research exploiting media data, 13th NII Testbeds Community for Information access Research (NTCIR-13) Medical document (MedWeb) provides pseudo-Twitter messages in a cross-language multi-label corpus, covering 3...

In the private residential sales market, obtaining orders for exterior design requires a proposal considering constraints of location and customer. The summary is document that describes concepts at earliest stage design. It not only appeals to customer, but also referenced in drawing. However, quality varies depending on skill creator because construction knowledge human experts. This paper aims generate using generative AI. Firstly, we analyze characteristics identify essential elements....

Profile inference of SNS users is valuable for marketing, target advertisement, and opinion polls.Several studies examining profile have been reported to date.Although information various types included in SNS, most such only use text information.It expected that incorporating other into classifiers can provide more accurate inference.As described this paper, we propose combined method processing image improve gender accuracy.By applying the simple formula combine two results derived from a...

Recommender systems aim to increase user actions such as clicks and purchases. Typical evaluations of recommenders regard the purchase a recommended item success. However, may have been purchased even without recommendation. An uplift is defined an in caused by recommendations. Situations with recommendation cannot both be observed for specific user-item pair at given time instance, making uplift-based evaluation optimization challenging. This paper proposes new metrics methods recommender...

Toru Nishino, Ryota Ozaki, Yohei Momoki, Tomoki Taniguchi, Ryuji Kano, Norihisa Nakano, Yuki Tagawa, Motoki Tomoko Ohkuma, Keigo Nakamura. Findings of the Association for Computational Linguistics: EMNLP 2020.

Recommender systems support user decision-making, and explanations of recommendations further facilitate their usefulness. Previous explanation styles are based on similar users, items, demographics contents items. Contexts, such as usage scenarios accompanying persons, have not been used for explanations, although they influence decisions. In this paper, we propose a context style method, presenting contexts suitable consuming recommended The expected impacts 1) persuasiveness: recognition...

We present a method for profiling businesses at specific locations that is based on mining information from social media. The matches geo-tagged tweets Twitter against venues Foursquare to identify the business mentioned in tweet. By linking geo-coordinates places, associated with business, such as store, can then be used profile business. sentiment estimator developed create profiles of stores chain, computing average each store. results heatmaps which show how differs across same chain and...

This paper describes our model for the reading comprehension task of MRQA shared task. We propose CLER, which stands Cross-task Learning with Expert Representation generalization and understanding. To generalize its capabilities, proposed is composed three key ideas: multi-task learning, mixture experts, ensemble. In-domain datasets are used to train validate model, other out-of-domain model’s performances. In a submission run result, achieved an average F1 score 66.1 % in setting, 4.3...

Demographic attribute inference of social networking service (SNS) users is a valuable application for marketing and targeting advertisements.Several studies have examined Twitter-user gender in natural language processing, image recognition, other research domains.Reportedly, combined approach using text data outperforms an individual approach.This paper presents proposal novel hybrid approach.A salient benefit our system that features provided from classifier are appropriately to infer...