Clustering of words with multiple meanings based on generating vectors for each meaning
A clustering method, clustering program, and clustering device are described herein for clustering of words with multiple meanings based on generating vectors for each meaning of a word. The generated vectors provide a distributed representation of a word in a vector space that account for the multiple meanings of the word so as, for instance, to learn semantic representations with high accuracy.
1. A non-transitory computer-readable recording medium having stored therein a clustering program for causing a computer to execute a process for generating a distributed representation of a word in a vector space, the process comprising:
recording, before a learning of multiple words, a first vector value in the vector space for each of the multiple words in at least one sentence retrieved from a sentence storage;
recording, after the learning, a second vector value in the vector space for each of the multiple words;
calculating a motion vector based on a difference between the first vector value and the second vector value for each of the multiple words;
clustering the multiple words based on the calculated motion vector; and
generating vectors separated for each meaning with respect to a word with multiple meanings included among the multiple words, based on a result of the clustering.
2. The recording medium according to claim 1 , wherein the at least one sentence includes multiple sentences, and the process further comprising:
setting a starting point in the vector space for a word included in multiple sentences;
treating each of the multiple sentences including the word as input, and acquiring a moved vector that has been moved from the starting point by learning;
computing a difference between the starting point when inputting each of the multiple sentences and the moved vector as a motion vector; and
clustering the multiple sentences based on a distance between the motion vectors for each of the multiple sentences.
3. The recording medium according to claim 2 , wherein the process further comprising:
assigning word labels to the word included in clustered sentences, in which a same label is assigned to a same cluster, and different labels are assigned to different clusters.
4. The recording medium according to claim 2 , wherein in the setting:
the starting point is set to a vector selected randomly in the vector space, to a zero vector whose elements are all zero, or to a vector obtained by learning according to a predetermined learning technique.
5. A clustering method, performed by a computer, for generating a distributed representation of a word in a vector space, the method comprising:
recording, before a learning of multiple words, a first vector value in the vector space for each of the multiple words in at least one sentence retrieved from a sentence storage;
recording, after the learning, a second vector value in the vector space for each of the multiple words;
calculating a motion vector based on a difference between the first vector value and the second vector value for each of the multiple words;
clustering the multiple words based on the calculated motion vector; and
generating vectors separated for each meaning with respect to a word with multiple meanings included among the multiple words, based on a result of the clustering.
6. A clustering apparatus for generating a distributed representation of a word in a vector space, the apparatus comprising:
a memory; and
a processor coupled to the memory and configured to perform a process including:
recording, before a learning of multiple words, a first vector value in the vector space for each of the multiple words in at least one sentence retrieved from a sentence storage;
recording, after the learning, a second vector value in the vector space for each of the multiple words;
calculating a motion vector based on a difference between the first vector value and the second vector value for each of the multiple words;
clustering the multiple words based on the calculated motion vector; and
generating vectors separated for each meaning with respect to a word with multiple meanings included among the multiple words, based on a result of the clustering.
7. A clustering apparatus for improving accuracy of a distributed representation of words, comprising:
an extractor unit configured to extract a word from sentences maintained in a sentence storage device and generate a listing of sentences including the word;
a first recorder unit configured to assign a first vector value in a vector space for the word before a learning of the word;
a second recorder unit configured to assign, after the learning, a second vector value in the vector space for the word based on a learning model for each of the sentences including the word;
a motion vector calculation unit configured to compute a motion vector based on a difference between the first vector value and each of the second vector values for each of the sentences including the word;
a sorter unit configured to cluster the word into one or more clusters based on the computed motion vectors;
a vector generating unit configured to generate one more vectors, each separated for a different meaning of word, to determine whether the word includes one or more meanings.