JPWO2021202552A5

JPWO2021202552A5 -

Info

Publication number: JPWO2021202552A5
Application number: JP2022559631A
Authority: JP
Publication date: 2023-10-12

Claims

A method,
receiving an utterance and a target domain for the chatbot;
generating a sentence embedding for the utterance;
obtaining an embedding representation for each cluster of a plurality of clusters of in-domain utterances associated with the target domain, wherein the embedding representation for each cluster is a sentence embedding for each in-domain utterance in the cluster. average, and the method further comprises:
inputting the sentence embedding for the utterance and the embedding representation for each cluster into a distance learning model, the distance learning model calculating a first probability as to whether the utterance belongs to the target domain; trained model parameters configured to provide, the method further comprising:
using the distance learning model to determine similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster;
Using the distance learning model, determine whether the utterance belongs to the target domain based on the determined similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster. predicting the first probability for
inputting the sentence embedding for the utterance and the embedding representation for each cluster into an outlier detection model, the outlier detection model being constructed with a distance or density algorithm for outlier detection. , the method further comprises:
using the outlier detection model to determine a distance or density deviation between the sentence embedding for the utterance and an embedding representation for an adjacent cluster;
predicting a second probability as to whether the utterance belongs to the target domain based on the determined distance or density deviation using the outlier detection model;
evaluating the first probability and the second probability to determine a final probability as to whether the utterance belongs to the target domain;
classifying the utterance as in-domain or out-of-domain for the chatbot based on the final probability.

Obtaining the embedded representation for each cluster comprises:
obtaining the in-domain utterance based on the target domain;
generating a sentence embedding for each in-domain utterance;
inputting the sentence embeddings for each in-domain utterance into an unsupervised clustering model, the unsupervised clustering model interpreting the in-domain utterance to configured to identify clusters of , and obtaining the embedded representation for each cluster further comprises:
Using the unsupervised clustering model, the sentence embeddings for each in-domain utterance are grouped into the plurality of clusters based on similarities and differences between the sentence embedding features and the sentence embedding features within each cluster. a step of classifying it into one of the
calculating a centroid for each cluster of the plurality of clusters;
2. The method of claim 1, comprising : outputting the embedded representation and the centroid for each cluster of the plurality of clusters.

calculating a z-score for the utterance based on the distance or density deviation between the sentence embedding for the utterance and the embedding representation for the adjacent cluster;
3. The method of claim 1 or 2 , further comprising determining the second probability as to whether the utterance belongs to the target domain by applying a sigmoid function to the z-score.

The sentence embedding for the utterance is generated using an embedding model that maps natural language elements, including sentences, words, and n-grams, to an array of numbers, where each of the natural language elements is a single digit in vector space. A method according to any one of claims 1 to 3 , expressed as points.

The step of determining the similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster includes (i) determining the similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster; (ii) inputting the absolute difference, the sentence embedding for the utterance, and the embedding representation for each cluster into a wide and deep learning network, The network comprises a linear model and a deep neural network, and the step of determining the similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster further comprises (iii) the linear model and the deep neural network. (iv) using the absolute difference to predict a wide-based probability as to whether the utterance belongs to the target domain; using the embedding representation to determine the similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster;
Predicting the first probability uses a final layer of the wide and deep learning network to predict the wide probability and the similarity between the sentence embedding for the utterance and each embedding representation for each cluster. The method according to any one of claims 1 to 4 , comprising the step of evaluating or evaluating differences.

the linear model comprises a plurality of model parameters trained using a set of training data;
The set of training data includes, for in-domain utterances from multiple domains, the absolute difference between a sentence embedding for an utterance and each embedding representation for each cluster;
during training of the linear model with the set of training data, using a hypothesis function to learn a linear relationship between the sentence embedding for the utterance and each embedding representation for each cluster;
6. The method of claim 5, wherein during learning the linear relationship, the plurality of model parameters are learned to minimize a loss function.

The wide and deep learning network comprises a plurality of model parameters trained using a set of training data;
The set of training data includes sentence embeddings for in-domain utterances from multiple domains;
During training of the wide and deep learning network with the set of training data, the high-dimensional features of the sentence embeddings for the in-domain utterances are transformed into low-dimensional vectors, which are then transformed into Concatenated with features from in-domain utterances and fed to a hidden layer of the deep neural network, the values of the low-dimensional vector are randomly initialized to minimize a loss function along with the plurality of model parameters. The method according to claim 5, wherein the method is trained as follows.

A computer program for causing one or more data processors to carry out the method according to any one of claims 1 to 7 .

A system,
one or more data processors;
a computer- readable storage medium containing instructions that, when executed on the one or more data processors, cause the one or more data processors to perform actions. and the action includes:
receiving target domains for utterances and chatbots;
generating a sentence embedding for the utterance;
obtaining an embedding representation for each cluster of a plurality of clusters of in-domain utterances associated with the target domain, wherein the embedding representation for each cluster includes a sentence embedding for each in-domain utterance in the cluster. average, and said action further includes:
inputting the sentence embedding for the utterance and the embedding representation for each cluster into a distance learning model, the distance learning model calculating a first probability as to whether the utterance belongs to the target domain. the learned model parameters configured to provide, the action further comprising:
using the distance learning model to determine similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster;
Using the distance learning model, determine whether the utterance belongs to the target domain based on the determined similarity or difference between the sentence embedding for the utterance and each embedding representation for each cluster. predicting the first probability for
inputting the sentence embedding for the utterance and the embedding representation for each cluster into an outlier detection model, the outlier detection model being constructed with a distance or density algorithm for outlier detection. , said action further includes:
using the outlier detection model to determine a distance or density deviation between the sentence embedding for the utterance and an embedding representation for an adjacent cluster;
using the outlier detection model to predict a second probability as to whether the utterance belongs to the target domain based on the determined distance or density deviation;
evaluating the first probability and the second probability to determine a final probability as to whether the utterance belongs to the target domain;
and classifying the utterance as in-domain or out-of-domain for the chatbot based on the final probability.