Learning device, learning method, computer program product, and information processing system

A learning method comprising:

• inputting, to an input layer of a neural network including M hidden layers and M output layers, one or more first morphemes conforming to any of M first arrangement patterns indicating an arrangement of one or more words, among morphemes included in a document, M being an integer, each of the M hidden layers being defined for one of the M first arrangement patterns, each of the M output layers being defined for one of the M first arrangement patterns, m-th (1≤m≤M) hidden layer among M hidden layers being connected to m-th output layer among M output layers and being connected to at least one of the M output layers other than the m-th output layer, each of the hidden layers being connected to the input layer without being connected to other hidden layers;
• learning the neural network to minimize a difference between one or more second morphemes conforming to any of second arrangement patterns indicating an arrangement of one or more words, among morphemes included in the document, and output morphemes from the neural network for the input first morphemes;
• and
• outputting an embedding vector of the first morphemes that is obtained based on a weight of the learned neural network.

A computer program product having a non-transitory computer readable medium including programmed instructions, wherein the instructions, when executed by a computer, cause the computer to perform:

• inputting, to an input layer of a neural network including M hidden layers and M output layers, one or more first morphemes conforming to any of M first arrangement patterns indicating an arrangement of one or more words, among morphemes included in a document, M being an integer, each of the M hidden layers being defined for one of the M first arrangement patterns, each of the M output layers being defined for one of the M first arrangement patterns, m-th (1≤m≤M) hidden layer among M hidden layers being connected to m-th output layer among M output layers and being connected to at least one of the M output layers other than the m-th output layer, each of the hidden layers being connected to the input layer without being connected to other hidden layers;
• learning the neural network to minimize a difference between one or more second morphemes conforming to any of a plurality of second arrangement patterns indicating an arrangement of one or more words, among morphemes included in the document, and output morphemes from the neural network for the input first morphemes;
• and
• outputting an embedding vector of the first morphemes that is obtained based on a weight of the learned neural network.

An information processing system comprising:

• one or more processors configured to
• input, to an input layer of a neural network including M hidden layers and M output layers, one or more first morphemes conforming to any of M first arrangement patterns indicating an arrangement of one or more words, among morphemes included in a document, M being an integer, each of the M hidden layers being defined for one of the M first arrangement patterns, each of the M output layers being defined for one of the M first arrangement patterns, m-th (1≤m≤M) hidden layer among M hidden layers being connected to m-th output layer among M output layers and being connected to at least one of the M output layers other than the m-th output layer, each of the hidden layers being connected to the input layer without being connected to other hidden layers, and
• learn the neural network to minimize a difference between one or more second morphemes conforming to any of second arrangement patterns indicating an arrangement of one or more words, among morphemes included in the document, and output morphemes from the neural network for the input first morphemes;
• and
• output an embedding vector of the first morphemes that is obtained based on a weight of the learned neural network.