Text2SQL can help non-professionals connect with databases by turning natural languages into SQL. Although previous researches about Text2SQL have provided some workable solutions, most of them extract values based on column representation. If there are multiple values in the query and these values belong to different columns, the previous approaches based on column representation cannot accurately extract values. In this work, we propose a new neural network architecture based on the pre-trained BERT, called M-SQL. The column-based value extraction is divided into two modules, value extraction and value-column matching. We evaluate M-SQL on a more complicated TableQA dataset, which comes from an AI competition. We rank first in this competition. Experimental results and competition ranking show that our proposed M-SQL achieves state-of-the-art results on TableQA.

It is of importance to model and estimate the user influence in social networks, especially for advertisers who conduct viral marketing. In this paper, we are interested in the number of received messages incurred by a node generating a message, and introduce the concepts of individual influence and type influence, while type influence is got by averaging out individual influence over nodes of the same type. We propose a user behavior model and use generating function to analyze type influence (including the mean and variance) and diffusion threshold, and find these results are not accurate in finite-size networks. We then classify nodes into subtypes and redefine the network model, which achieves much more accurate results. We also propose a scalable approach to estimate individual influence, and find it can get good approximates for individual influence, subtype influence and type influence by only considering local neighbors and out-of-date information, which is useful in large-scale networks. All analysis results are verified by simulations in real-world networks. Models in this paper can be extended to consider more realistic situations, and we believe these results are of use in understanding the diffusion dynamics in social networks.

Automatically parsing SQL queries from natural languages can help non-professionals access databases and improve the efficiency of information utilization. It is a long-term research issue and has recently received attention from the relevant communities. Although previous researches have provided some workable solutions, most of them only consider table schemas and natural language questions when parsing SQL queries, and do not use table contents. We observe that table contents can provide more helpful information for some user questions. In this paper, we propose a novel neural network approach, F-SQL, to focus on solving the problem of table content utilization. In particular, we employ the gate mechanism to fuse table schemas and table contents and get the more different representation about table schemas. We test this idea on the WikiSQL and TableQA datasets. Experimental results show that F-SQL achieves new state-of-the-art results on WikiSQL and TableQA.

In the Internet era, online clustering of technology web news can help discover scientific breakthroughs and grasp technology trends. To do that automatically, the news documents to be clustered must be represented appropriately with numerical vectors. However, traditional representations such as Term Frequency-Inverse Document Frequency (TF-IDF) cannot distinguish near-synonyms and may cause “dimension disaster.” To overcome these problems, this article proposes the Bag-of-Near-Synonyms (BoNS) model based on the idea to construct near-synonym sets using word embeddings and agglomerative clustering, and then to represent a document with a Set Frequency-Inverse Document Frequency (SF-IDF) vector in which each dimension corresponds to a near-synonym set rather than a single word. To speed up computation, we further propose the hashed version of SF-IDF and name it hSF-IDF, which employs a hash function to map each near-synonym set to a unique number as the key and hence reduces the computation of SF to linear time. In addition, we apply hSF-IDF to online clustering of Chinese technology web news and propose an improved batch-based method. Extensive experiments have been conducted on a real-world dataset. The results show that our model outperforms some strong baselines including TF-IDF, average pooling of word or character embeddings, Latent Dirichlet Allocation (LDA), and bag-of-concepts in terms of both accuracy and efficiency.

Bipartite network projection method has been recently employed for personal recommendation. It constructs a bipartite network between users and items. Treating user taste for items as resource in the network, we allocate the resource via links between user nodes and item nodes. However, the taste model employed by existing algorithms cannot differentiate “dislike” and “unrated” cases implied by user ratings. Moreover, the distribution of resource is solely based on node degrees, ignoring the different transfer rates of the links. To enhance the performance, this paper devises a negative‐aware and rating‐integrated algorithm on top of the baseline algorithm. It enriches the current user taste model to encompass “like,” “dislike,” and “unrated” information from users. Furthermore, in the resource distribution stage, we propose to initialize the resource allocation according to user ratings, which also determines the resource transfer rates on links afterward. Additionally, we also present a scalable implementation in the MapReduce framework by parallelizing the algorithm. Extensive experiments conducted on real data validate the effectiveness and efficiency of the proposed algorithms.

Recommendation is increasingly widely used in many areas in which collaborative filtering is one of the most well-known algorithms, but the traditional collaborative filtering (CF) recommendation algorithm suffers from data sparsity and user similarity imprecision problems. To conquer these problems mentioned above, this paper studies user attributes representing and proposes a hybrid recommendation algorithm based on attribute knowledge embedded collaborative filtering. By making use of attribute information, the similarity between users is improved. The hybrid algorithm assembles user attribute and collaborative filtering by a weighted method. Experiments validates its effectiveness.