Information processing apparatus, information processing method, and non-transitory computer readable recording medium
An information processing apparatus according to the present application includes a classification unit, a conversion unit, a label acquiring unit. The classification unit that classifies a feature vector by using a classification vector, the feature vector being converted from target data. The conversion unit that converts the feature vector into an embedding vector in accordance with a conversion rule according to classification performed by the classification unit. The label acquiring unit that acquires, as one or more labels to be provided to the target data, one or more labels acquired based on the embedding vector converted by the conversion unit.
1. An information processing apparatus comprising:
a processor programmed to:
classify a feature vector by using a classification vector, the feature vector being converted from user inputted target data, which includes at least one of a search query and a browsing activity of the user;
convert the classified feature vector into an embedding vector in accordance with a conversion rule;
acquire, as one or more labels to be provided to the target data, one or more labels acquired based on the converted embedding vector;
acquire (i) a first feature vector corresponding to a first label vector and
(ii) a predetermined number of second feature vectors corresponding to a predetermined number of vectors of second label vectors, values of inner products between the first label vector and the predetermined number of second label vectors being upper-order values; and
learn the classification vector by using the first feature vector and the predetermined number of second feature vectors as learning data.
2. The information processing apparatus according to claim 1 , wherein the processor is programmed to:
acquire an embedding matrix in accordance with the classification; and
multiply the feature vector by the acquired embedding matrix to compute an embedding vector.
3. The information processing apparatus according to claim 2 , further comprising a memory configured to store a plurality of label data obtained by associating label vectors with embedding vectors, wherein:
the processor is programmed to:
search, by using an approximate nearest neighbor search, the plurality of label data stored in the memory for one or more embedding vectors similar to the computed embedding vector,
acquire one or more label vectors associated with the searched one or more embedding vectors, and
acquire one or more labels corresponding to the acquired one or more label vectors as the one or more labels to be provided to the target data.
4. The information processing apparatus according to claim 1 , wherein the processor is programmed to adjust the classification vector to classify the first feature vector so that a cosine similarity between the classification vector and the predetermined number of second feature vectors is large.
5. The information processing apparatus according to claim 4 , wherein the processor is programmed to adjust the classification vector to classify the first feature vector so that a cosine similarity between the classification vector and one or more third feature vectors acquired at random is small.
6. An information processing apparatus comprising: a processor programmed to:
classify a feature vector by using a classification vector, the feature vector being converted from user inputted target data, which includes at least one of a search query and a browsing activity of the user;
convert the classified feature vector into an embedding vector in accordance with a conversion rule;
acquire, as one or more labels to be provided to the target data, one or more labels acquired based on the converted embedding vector;
acquire (i) a first embedding vector corresponding to a first label vector, and
(ii) a predetermined number of second embedding vectors corresponding to a predetermined number of vectors of second label vectors; values of inner products between the first label vector and the predetermined number of second label vectors being upper-order values; and
learn the embedding matrix by using the first embedding vector and the predetermined number of second embedding vectors as learning data.
7. The information processing apparatus according to claim 6 , wherein the processor is programmed to adjust the embedding matrix so that a cosine similarity between the first embedding vector and the predetermined number of second embedding vectors is large.
8. The information processing apparatus according to claim 7 , wherein the processor is programmed to adjust the embedding matrix so that a cosine similarity between the first embedding vector and one or more third embedding vectors acquired at random is small.
9. An information processing method comprising:
classify a feature vector by using a classification vector, the feature vector being converted from user inputted target data, which includes at least one of a search query and a browsing activity of the user;
converting the classified feature vector into an embedding vector in accordance with a conversion-rule;
acquiring, as one or more labels to be provided to the target data, one or more labels acquired based on the converted embedding vector;
acquiring (i) a first feature vector corresponding to a first label vector and
(ii) a predetermined number of second feature vectors corresponding to a predetermined number of vectors of second label vectors, values of inner products between the first label vector and the predetermined number of second label vectors being upper-order values; and
learning the classification vector by using the first feature vector and the predetermined number of second feature vectors as learning data.
10. A non-transitory computer-readable recording medium having stored therein a program that causes a computer to execute a process comprising:
classify a feature vector by using a classification vector, the feature vector being converted from user inputted target data, which includes at least one of a search query and a browsing activity of the user;
converting the classified feature vector into an embedding vector in accordance with a conversion rule;
acquiring, as one or more labels to be provided to the target data, one or more labels acquired based on the converted embedding vector;
acquiring (i) a first feature vector corresponding to a first label vector and
(ii) a predetermined number of second feature vectors corresponding to a predetermined number of vectors of second label vectors, values of inner products between the first label vector and the predetermined number of second label vectors being upper-order values; and
learning the classification vector by using the first feature vector and the predetermined number of second feature vectors as learning data.