Method and device for extracting date and time from human-computer interaction text

Abstract

The invention discloses a method and a device for extracting date and time from a human-computer interaction text. The method comprises the following steps: S101, segmenting a text into a plurality oftext segments; S 102, calculating semantic similarity between the intention text and each segmented text segment; S 103, comparing the semantic similarity of each segmented text segment with a threshold value; and S104, if the semantic similarity of any segmented text segment in the segmented text segments is greater than a threshold value, standardizing the time text contained in any segmented text segment with a timestamp, generating a response, and returning the response. The invention discloses a method for extracting date and time from a human-computer interaction text. In an outbound multi-round dialogue scene, a time label is marked on a speaking text of a user through NER in NLP, then the intention expressed by the user is matched through a semantic similarity model, and the timelabel and the intention are combined, so that the business time slot extraction accuracy can be well improved, and the interactive experience of a product is improved.

Description

technical field [0001] The present invention relates to the field of human-computer interaction technology, in particular to a method and device for extracting date and time from human-computer interaction text. Background technique [0002] In the multi-round interaction scenario of outbound calls, the time information of the user in a certain round of interaction needs to be extracted due to business needs. For example, in the collection scenario, the specific repayment time of the user needs to be extracted so that the business can judge whether there is a risk of overdue. Through the model training of the time dictionary, use the named entity recognition (NER) to label the time text. If there are multiple times, use the time label as a separator to divide the sentence, and then use the acquaintance model For similarity calculation, you can select an initial threshold of 0.8 (which can be configured in the cloud according to the actual test value). If a keyword is matched...

AI Technical Summary

Problems solved by technology

[0003] Mark the time tag through the named entity recognition (NER) in natural language processing (NLP), and then extract the text of the time from the text in the conversation in the user. If the text spoken by the user contains multiple time tags, the time information extracted in this way , not only cannot effectively distinguish whether it is the time required by the business, but also seriously affects the user experience, and even causes user complaints

Images