Search

Intelligent finance is a new form of business with deep integration of artificial intelligence technology and financial industry. An important application of intelligent finance is the precise marketing of financial products. As a key link in precision marketing, click through rate(CTR) prediction has made great progress, but there is still room for improvement in multiple features fusion, feature interactions learning and other aspects. In view of these needs and challenges, we propose a CTR prediction model named MCGM, which is used to realize precision marketing of financial products. The main characteristics of the model are as follows: (i) in order to effectively fuse multiple features, we design a hierarchical gated mechanism to select salient feature information at different levels; (ii) in order to fully learn the nonlinear relationship between features, we design a multi-channel feature interactions learning module. Specifically, it adopts factorization machine(FM), improved CrossNet(ICN) and multilayer perceptron(MLP) components to model the feature interactions from high-order to low-order, in order to obtain the abstract features containing rich information. Comprehensive and sufficient experiments on real world datasets show that the proposed model achieves better prediction performance compared with baselines. The proposed model not only has specific application value in the field of financial products marketing, but also provides an idea reference for data-driven marketing modeling. [ABSTRACT FROM AUTHOR]

Every epidemic affects the real lives of many people around the world and leads to terrible consequences. Recently, many tweets about the COVID-19 pandemic have been shared publicly on social media platforms. The analysis of these tweets is helpful for emergency response organizations to prioritize their tasks and make better decisions. However, most of these tweets are non-informative, which is a challenge for establishing an automated system to detect useful information in social media. Furthermore, existing methods ignore unlabeled data and topic background knowledge, which can provide additional semantic information. In this paper, we propose a novel Topic-Aware BERT (TABERT) model to solve the above challenges. TABERT first leverages a topic model to extract the latent topics of tweets. Secondly, a flexible framework is used to combine topic information with the output of BERT. Finally, we adopt adversarial training to achieve semi-supervised learning, and a large amount of unlabeled data can be used to improve inner representations of the model. Experimental results on the dataset of COVID-19 English tweets show that our model outperforms classic and state-of-the-art baselines. [ABSTRACT FROM AUTHOR]

With the rapid development of artificial intelligence and the increasing demand for legal intelligence, using AI methods to predict legal judgments has become a hot spot in recent years. Charge prediction is one of the core tasks of Legal Judgment Prediction (LJP). It aims to predict charge from complicated legal facts, so as to help the court make judgments or provide legal professional guidance to non-professionals. In the field of legalAI, interpretability is crucial compared to others. Reasonable interpretability can eliminate hidden dangers such as gender discrimination and provide support for judges' decisions. However, how to add the legal theory framework to the modeling to improve the interpretability is a challenge, which has few researches at present. To address this problem, we use Double-layer Criminal System as a guide to build Charge Prediction modeling called DCSCP which aims to predict charges in the criminal law of China. In general, our characteristic is to achieve multi-granularity inference of legal charges by obtaining the subjective and objective elements from the fact descriptions of legal cases. Specifically, our approach is performed in two steps: (1) extract the objective elements from the fact description and use them to generate candidate charges to achieve coarse-grained prediction; (2) extract the subjective elements from the fact description, and design the first-order predicate logic inference to realize the fine-grained charge inference in combination with the candidate charges. Experimental results show that our DCSCP can provide interpretable predictions, and it can maintain performance compared to other state-of-the-art charge prediction models. [ABSTRACT FROM AUTHOR]

Cross-lingual Open Domain Question Answering (Cross-lingual Open-QA) has been developed since it was proposed in the mid-1990s. It can be divided into two mainstream tasks according to the training corpus used in the answer extraction stage. One is that both of the training and testing data are in the target language. The other is that the training data is in the source language, and the testing data is in the target language. For a long time, the former has been studied mainly through translation based approaches. Until 2019, the latter appeared and non-translation based approaches become available thanks to multilingual BERT model. Therefore, the two tasks have been discussed separately, which encourages our work on whether it is possible to achieve these two tasks simultaneously without any additional transformation. It is observed that the existence of the multilingual BERT model makes a solution to establish a unified framework. However, there are two problems with using the multilingual BERT model directly. The one is in the document retrieval stage, directly working multilingual pretraining model for similarity calculation will result in insufficient retrieval accuracy. The other is in the answer extraction stage, the answers will involve different levels of abstraction related to retrieved documents, which needs deep exploration. This paper puts forward a multi-granularity semantic space learning based approach for cross-lingual Open-QA. It consists of the Match-Retrieval module and the Multi-granularity-Extraction module. The matching network in the retrieval module makes heuristic adjustment and expansion on the learned features to improve the retrieval quality. In the answer extraction module, the reuse of deep semantic features is realized at the network structure level through cross-layer concatenation, and it enables us to learn multi-granularity semantic space. Experimental results on two public cross-lingual Open-QA datasets show the superiority of our proposed approach over the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

With the rapid development of knowledge bases (KBs), question answering (QA) based on KBs has become a hot research issue. The KB-QA technology can be divided into two technical routes: (1) symbol based representations, such as traditional semantic parsing, and (2) distribution based embedding. With the emergence of deep learning, the development of NLP has greatly promoted. The effect of KB-QA can be improved by combining deep learning with the above two technical routes respectively. In this paper, the impact of the second route (i.e., Distribution Embedding) combined with deep learning is mainly discussed. This route can be divided into pipeline frameworks and end-to-end frameworks. For comprehensive analysis, two frameworks (i.e., a pipeline framework, an end-to-end framework) are proposed to focus on answering single-relation factoid questions. In both of two frameworks, the effect of context information on the quality of QA is studied, such as the entity's notable type, out-degree. In the pipeline framework, it includes two cascaded steps: entity detection and relation detection. In this framework, multiple modules need to be built, and corresponding training data sets must be constructed for them respectively. The entire process of the pipleine framework is complex, costly and has the problem of error propagation. In the end-to-end framework, the two subtasks of entity detection and relation detection are merged together, and then combined into one framework. Questions, entities and relations are mapped into the same semantic space through the encoding of the recurrent neural network. Moreover, the question-entity similarity and the question-relation similarity are calculated, so that the candidate answers can be sorted and selected. Moreover, character-level(char-level) encoding and self-attention mechanisms are combined using weight sharing and multi-task strategies to enhance the accuracy of QA. Experimental results show that context information can get better results of simple QA whether it is the pipeline framework or the end-to-end framework. In addition, the end-to-end framework achieves results competitive with state-of-the-art approaches in terms of accuracy. [ABSTRACT FROM AUTHOR]

Keyword-based Web search is a widely used approach for locating information on the Web. However, Web users usually suffer from the difficulties of organizing and formulating appropriate input queries due to the lack of sufficient domain knowledge, which greatly affects the search performance. An effective tool to meet the information needs of a search engine user is to suggest Web queries that are topically related to their initial inquiry. Accurately computing query-to-query similarity scores is a key to improve the quality of these suggestions. Because of the short lengths of queries, traditional pseudo-relevance or implicit-relevance based approaches expand the expression of the queries for the similarity computation. They explicitly use a search engine as a complementary source and directly extract additional features (such as terms or URLs) from the top-listed or clicked search results. In this paper, we propose a novel approach by utilizing the hidden topic as an expandable feature. This has two steps. In the offline model-learning step, a hidden topic model is trained, and for each candidate query, its posterior distribution over the hidden topic space is determined to re-express the query instead of the lexical expression. In the online query suggestion step, after inferring the topic distribution for an input query in a similar way, we then calculate the similarity between candidate queries and the input query in terms of their corresponding topic distributions; and produce a suggestion list of candidate queries based on the similarity scores. Our experimental results on two real data sets show that the hidden topic based suggestion is much more efficient than the traditional term or URL based approach, and is effective in finding topically related queries for suggestion. [ABSTRACT FROM AUTHOR]

In recent years, lots of knowledge graphs built from Wikipedia, the largest multilingual online encyclopedia, have been published on the Web to support various applications. However, since non-English data in Wikipedia are sparse, some projects work on knowledge graph construction from multiple non-English online encyclopedias, but many technical details are missing, so it is hard to reuse their frameworks or techniques. In this paper, we propose a new framework to solve knowledge graph construction from multiple online encyclopedias. The core modules are knowledge extraction and knowledge linking. Knowledge extraction consists of regular extraction, i.e., extracting targeted article contents in the whole online encyclopedias periodically, and live extraction, which only extracts the article contents of new and updated entities. Knowledge linking utilizes heuristic lightweight entity matching strategies and a semi-supervised learning method to find duplicated entities and properties from different online encyclopedias. Experimental results show that our approaches for knowledge extraction and linking outperform state-of-the-art baselines in different evaluation metrics, and our framework can generate a large-scale knowledge graph after inputting multiple online encyclopedias. [ABSTRACT FROM AUTHOR]

With the popularization of social media and the exponential growth of information generated by online users, the recommender system has been popular in helping users to find the desired resources from vast amounts of data. However, the cold-start problem is one of the major challenges for personalized recommendation. In this work, we utilized the tag information associated with different resources, and proposed a tag-based interactive framework to make the resource recommendation for different users. During the interaction, the most effective tag information will be selected for users to choose, and the approach considers the users’ feedback to dynamically adjusts the recommended candidates during the recommendation process. Furthermore, to effectively explore the user preference and resource characteristics, we analyzed the tag information of different resources to represent the user and resource features, considering the users’ personal operations and time factor, based on which we can identify the similar users and resource items. Probabilistic matrix factorization is employed in our work to overcome the rating sparsity, which is enhanced by embedding the similar user and resource information. The experiments on real-world datasets demonstrate that the proposed algorithm can get more accurate predictions and higher recommendation efficiency. [ABSTRACT FROM AUTHOR]

This work proposes a novel data clustering algorithm based on the potential field model, with a hierarchical optimization mechanism on the algorithm. There are two stages in this algorithm. Firstly, we build an edge-weighted tree based on the mutual distances between all data points and their hypothetical potential values derived from the data distribution. Using the tree structure, the dataset can be divided into an appropriate number of initial sub-clusters, with the cluster centers close to the local minima of the potential field. Then the sub-clusters are further merged according to the well-designed merging criteria by analyzing their border potential values and the cluster average potential values. The proposed clustering algorithm follows a hierarchical clustering mechanism, and aims to optimize the initial sub-cluster results in the first stage. The algorithm takes advantage of the cluster merging criteria to merge the sub-clusters, so it can automatically stop the clustering process without designating the number of clusters in advance. The experimental results show that the proposed algorithm produces the most satisfactory clustering results in most cases compared with other existing methods, and can effectively identify the data clusters with arbitrary shape, size and density. [ABSTRACT FROM AUTHOR]