https://www.socher.org/index.php/Main/ReasoningWithNeuralTensorNetworksForKnowledgeBaseCompletion

Year: 2013

https://www.cnblogs.com/wuseguang/p/4168963.html

https://blog.csdn.net/wty__/ article/details/52447128

Socher et al. Proposed rntn (recurrent neural tensor networks) model in 2013, which uses tensor to represent the combination parameters. The commonly used third-order tensor can be understood as a vector composed of multiple matrices, in which each matrix can be considered as some kind of combination operation, and finally combined together. Through tensor, we can not only reduce the parameters needed to learn, but also express rich combination operations, so rntn model also achieves good results. In the fine-grained sentiment analysis task, the accuracy of mv-rnn is improved from 44.4% to 45.7%. Among them, fine-grained emotion classification refers to the classification of the emotions expressed in sentences, such as commendation and derogation, into five levels, corresponding to the star rating of goods by the review website

Knowledge base:
openkg collects and collates important open knowledge base and knowledge mapping projects at home and abroad, and organizes and collates relevant Chinese materials, which are open to the public free of charge. Yago is a linked database developed by Max Planck Institute in Germany. Yago mainly integrates data from Wikipedia, WordNet and GeoNames. Yago integrates the vocabulary definition of WordNet with the classification system of Wikipedia, which makes Yago have a richer entity classification system. Yago also considers temporal and spatial knowledge, and adds attribute description of temporal and spatial dimensions for many knowledge items. At present, Yago contains 120 million pieces of triple knowledge. Yago is one of the back-end knowledge bases of IBM Watson

Hadamard product: Hadamard product of m x n matrix A = [AIJ] and matrix B = [bij], denoted as a * B. The new matrix element is defined as the product (a * b) ij = AIJ * bij of the corresponding elements of matrix A and B
identity: identity

Contribution:
1) the role of the model is to predict other facts on the basis of known facts. For example, when someone tells you that a new monkey has just been discovered, you don’t need to look for evidence to know that the monkey also has legs 2) this paper introduces a new method to represent entities in the knowledge base, we represent each entity as the average of its word vectors, allowing the sharing of statistical strength between the words describing each entity e.g, Bank of China and China.
in previous work, each entit was represented by a vector, The incorporation of word vectors which are trained on large unlabeled text. This ready available resource enables all models to more accurately predict relationships. This readily available resource enables all models to predict relationships more accurately

Understanding bilinear

It’s not very relevant to my project