Dynamic Relevance Graph Network for Knowledge-Aware Question Answering.

1. The research deals with the challenge of learning and reasoning for Commonsense Question Answering given an external source of knowledge in the form of a knowledge graph.
2. We propose a novel graph neural network architecture, called Dynamic Relevance Graph Network (DRGN).
3. DRGN operates on a given KG subgraph based on the question and answers entities and uses the relevance scores between the nodes to establish new edges dynamically for learning node representations in the graph network.
4. The model exploits the existing relationships but re-scales the weights and influences the way the neighborhood nodes' representations are aggregated in the KG subgraph.
5. The model potentially recovers the missing edges in KG that are needed for the chain of reasoning.
6. Competitive performance on CommmonsenseQA and OpenbookQA benchmarks.
7. The paper published in COLING 2022.
8. Paper name: Dynamic Relevance Graph Network for Knowledge-Aware Question Answering.

Relevant CommonSense Subgraphs for "What if..." Procedural Reasoning.

1. The research deals with the challenge of learning causal reasoning over procedural text to answer "What if..." questions when external commonsense knowledge is required.
2. We propose a novel multi-hop graph reasoning model to extract a commonsense subgraph with the most relevant information from a large knowledge graph.
3. The model predicts the causal answer by reasoning over the representations obtained from the commonsense subgraph and the contextual interactions between the questions and context.
4. Competitive performance on WIQA tasks.
5. The paper published in ACL 2022 findings.
6. Paper name: Relevant CommonSense Subgraphs for "What if..." Procedural Reasoning.

Relational Gating for "What If" Reasoning.

1. The research deals with the challenge of learning and reasoning over multi-hop "What If" question answering (QA).
2. We propose entity gating and relational gating mechanism to capture the most important entities and relationships involved in qualitative comparison, causal reasoning and multi-hop reasoning.
3. We propose a contextual interaction module to effectively and efficiently align the question and paragraph entities.
4. Our proposed approach shows competitive performance on the WIQA benchmark.
5. The paper published in IJCAI 2021.
6. Paper name: Relational Gating for "What If" Reasoning.

SRLGRN: Semantic Role Labeling Graph Reasoning Network.

1. The research deals with the challenge of learning and reasoning over multi-hop question answering (QA).
2. We build a heterogeneous semantic role labeling graphs.
3. We propose a graph reasoning network based on the semantic structure of the sentences.
4. The model learns cross paragraph reasoning paths and find the supporting facts and the answer jointly.
5. Our proposed approach shows competitive performance on the HotpotQA distractor setting benchmark.
6. The paper published in EMNLP 2020.
7. Paper name: SRLGRN: Semantic Role Labeling Graph Reasoning Network.

Cross-Modality Relevance for Reasoning on Language and Vision.

1. Designing a novel cross-modality relevance model to learn the relevance representation between components of various input modalities.
2. The model includes the higher-order relevance between entity relations in the text and object relations in the image.
3. Dataset: NLVR2 and VQA.
4. Competitive performance on above two language and vision tasks.
5. The paper published in ACL 2020.
6. Paper name: Cross-Modality Relevance for Reasoning on Language and Vision.

Internship in Big Data and Information Retrieval group, JD Inc.

1. Our IR team presented a novel approach called DPSR, which stands for Deep Personalized and Semantic Retrieval.
2. During the internship, I helped the team design and implement the item module of the DPSR retrieval system at the early stage of this project.
3. Dataset: The DPSR model is trained on a data set of 60 days JD user click logs, which contains 5.6 billion sessions.
4. State-of-the-art performance in Offline test (Top-K and AUC) and Online A/B test (UCVR, GMV, QRR).
5. The paper published in SIGIR 2020. My name appears in Acknowledgement section.
6. Paper name: Towards Personalized and Semantic Retrieval: An End-to-End Solution for E-commerce Search via Embedding Learning.

Internship in NLP group, Baidu Inc.

1. Using Deep Learning based semantic matching method to solve Learn-to-Rank problem.
2. Dataset: Baidu Clickthrough dataset.
3. State-of-the-art performance in NDCG and MAP.
4. Paper published in IJCAI 2019.
5. Paper name: RLTM: An Efficient Neural IR Framework for Long Documents.

A Deep Learning Approach for Expert Identification in Question Answering Communities.

1. Building up a language model to implement expert identification in QA communities.
2. Natural Language Processing technologies, such as Word2vec, Glove, DeepWalk, and some Deep Learning technologies, such as Convolutional neural network, Recurrent neural network.
3. Dataset: Stack Overflow community, Zhihu question-answering community.
4. The top-1 test accuracy outperforms all of the baselines.
5. Paper published in arXiv: 1711.05350.

Design and Development of Pet Shop Trade System Based on Java Web.

1. Front-end Design and implement with HTML/CSS, JavaScript, JQuery.
2. Backend Design and implementation with Struts2, Hibernate, Spring3.
3. Relational Database: MySQL, NoSql Database: Redis
4. Full-Stack design and implementation of recommendation system(Mahout) and search engine(Lucene).
5. Introduced open platform, which can log into the account via Facebook and Tencent account.