A5044042138- Topic Modeling
- Natural Language Processing Techniques
- Speech and dialogue systems
- Multimodal Machine Learning Applications
- Advanced Text Analysis Techniques
- Sentiment Analysis and Opinion Mining
- Advanced Graph Neural Networks
- Adversarial Robustness in Machine Learning
- Biomedical Text Mining and Ontologies
- Text and Document Classification Technologies
- Mental Health via Writing
- AI in Service Interactions
- Semantic Web and Ontologies
- Speech Recognition and Synthesis
- Hate Speech and Cyberbullying Detection
- Data Quality and Management
- Computational and Text Analysis Methods
- Recommender Systems and Techniques
- Artificial Intelligence in Games
- Multi-Agent Systems and Negotiation
- Domain Adaptation and Few-Shot Learning
- Machine Learning in Healthcare
- Explainable Artificial Intelligence (XAI)
- Text Readability and Simplification
- Digital Mental Health Interventions
Tsinghua University
2016-2025
IT University of Copenhagen
2023
Tokyo Institute of Technology
2023
Administration for Community Living
2023
National Engineering Research Center for Information Technology in Agriculture
2018-2023
Intelligent Health (United Kingdom)
2016-2023
Center for Information Technology
2019-2023
American Jewish Committee
2023
University of Michigan
2019-2023
China Mobile (China)
2022
Aspect-level sentiment classification is a finegrained task in analysis.Since it provides more complete and in-depth results, aspect-level analysis has received much attention these years.In this paper, we reveal that the polarity of sentence not only determined by content but also highly related to concerned aspect.For instance, "The appetizers are ok, service slow.",for aspect taste, positive while for service, negative.Therefore, worthwhile explore connection between an sentence.To end,...
Large-scale pre-trained models (PTMs) such as BERT and GPT have recently achieved great success become a milestone in the field of artificial intelligence (AI). Owing to sophisticated pre-training objectives huge model parameters, large-scale PTMs can effectively capture knowledge from massive labeled unlabeled data. By storing into parameters fine-tuning on specific tasks, rich implicitly encoded benefit variety downstream which has been extensively demonstrated via experimental...
Commonsense knowledge is vital to many natural language processing tasks. In this paper, we present a novel open-domain conversation generation model demonstrate how large-scale commonsense can facilitate understanding and generation. Given user post, the retrieves relevant graphs from base then encodes with static graph attention mechanism, which augments semantic information of post thus supports better post. Then, during word generation, attentively reads retrieved triples within each...
Recently, knowledge graph embedding, which projects symbolic entities and relations into continuous vector space, has become a new, hot topic in artificial intelligence.This paper proposes novel generative model (TransG) to address the issue of multiple relation semantics that may have meanings revealed by entity pairs associated with corresponding triples.The new can discover latent for leverage mixture relationspecific component vectors embed fact triple.To best our knowledge, this is...
In the past few years, sentiment analysis and opinion mining becomes a popular important task. These studies all assume that their resources are real trustful. However, they may encounter faked or spam problem. this paper, we study issue in context of our product review system. On site, people write reviews, called spam, to promote products, defame competitors' products. It is identify filter out spam. Previous work only focuses on some heuristic rules, such as helpfulness voting, rating...
Building dialogue systems that can converse naturally with humans is a challenging yet intriguing problem of artificial intelligence. In open-domain human-computer conversation, where the conversational agent expected to respond human utterances in an interesting and engaging way, commonsense knowledge has be integrated into model effectively. this paper, we investigate impact providing about concepts covered dialogue. Our represents first attempt integrating large base end-to-end models....
Existing relation classification methods that rely on distant supervision assume a bag of sentences mentioning an entity pair are all describing for the pair. Such methods, performing at level, cannot identify mapping between and sentence, largely suffers from noisy labeling problem. In this paper, we propose novel model sentence level data. The has two modules: instance selector classifier. chooses high-quality with reinforcement learning feeds selected into classifier, classifier makes...
There is a resurgent interest in developing intelligent open-domain dialog systems due to the availability of large amounts conversational data and recent progress on neural approaches AI [33]. Unlike traditional task-oriented bots, an system aims establish long-term connections with users by satisfying human need for communication, affection, social belonging. This article reviews work that are devoted addressing three challenges such systems: semantics , consistency interactiveness ....
Story generation, namely, generating a reasonable story from leading context, is an important but challenging task. In spite of the success in modeling fluency and local coherence, existing neural language generation models (e.g., GPT-2) still suffer repetition, logic conflicts, lack long-range coherence generated stories. We conjecture that this because difficulty associating relevant commonsense knowledge, understanding causal relationships, planning entities events with proper temporal...
Most existing methods determine relation types only after all the entities have been recognized, thus interaction between and entity mentions is not fully modeled. This paper presents a novel paradigm to deal with extraction by regarding related as arguments of relation. We apply hierarchical reinforcement learning (HRL) framework in this enhance types. The whole process decomposed into hierarchy two-level RL policies for detection respectively, so that it more feasible natural overlapping...
Generating a reasonable ending for given story context, i.e., generation, is strong indication of comprehension. This task requires not only to understand the context clues which play an important role in planning plot, but also handle implicit knowledge make reasonable, coherent story. In this paper, we devise novel model generation. The adopts incremental encoding scheme represent are spanning context. addition, commonsense applied through multi-source attention facilitate comprehension,...
Text classification is a fundamental problem in natural language processing. As popular deep learning model, convolutional neural network (CNN) has demonstrated great success this task. However, most existing CNN models apply convolution filters of fixed window size, thereby unable to learn variable n-gram features flexibly. In paper, we present densely connected with multi-scale feature attention for text classification. The dense connections build short-cut paths between upstream and...
Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, Minlie Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.
We introduce ChatGLM, an evolving family of large language models that we have been developing over time. This report primarily focuses on the GLM-4 series, which includes GLM-4, GLM-4-Air, and GLM-4-9B. They represent our most capable are trained with all insights lessons gained from preceding three generations ChatGLM. To date, pre-trained ten trillions tokens mostly in Chinese English, along a small set corpus 24 languages, aligned for English usage. The high-quality alignment is achieved...
With the development of Web 2.0, sentiment analysis has now become a popular research problem to tackle. Recently, topic models have been introduced for simultaneous topics and in document. These studies, which jointly model sentiment, take advantage relationship between are shown be superior traditional tools. However, most them make assumption that, given parameters, sentiments words document all independent. In our observation, contrast, expressed coherent way. The local conjunctive...
Abstract Motivation: Although there are several databases storing protein–protein interactions, most such data still exist only in the scientific literature. They scattered literature written natural languages, defying mining efforts. Much time and labor have to be spent on extracting protein pathways from Our aim is develop a robust powerful methodology mine interactions biomedical texts. Results: We present novel approach for method uses dynamic programming algorithm compute distinguishing...
We report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers genes detected full-text articles. For training, 32 fully and 500 partially annotated articles prepared. A total 507 selected as test set. Due high annotation cost, it was not feasible obtain gold-standard human annotations for all Instead, we developed an Expectation Maximization (EM) algorithm approach choosing small number manual that most capable...
This paper deals with sentence-level sentiment classification. Though a variety of neural network models have been proposed recently, however, previous either depend on expensive phrase-level annotation, most which has remarkably degraded performance when trained only annotation; or do not fully employ linguistic resources (e.g., lexicons, negation words, intensity words). In this paper, we propose simple but also attempt to model the role and words. Results show that our are able capture...
Existing relation classification methods that rely on distant supervision assume a bag of sentences mentioning an entity pair are all describing for the pair. Such methods, performing at level, cannot identify mapping between and sentence, largely suffers from noisy labeling problem. In this paper, we propose novel model sentence level data. The has two modules: instance selector classifier. chooses high-quality with reinforcement learning feeds selected into classifier, classifier makes...
Due to the high cost of manual curation key aspects from scientific literature, automated methods for assisting this process are greatly desired. Here, we report a novel approach facilitate MeSH indexing, challenging task assigning terms MEDLINE citations their archiving and retrieval.