Search

CN-121996795-A - Method and system for classifying telephone voices in hotel

CN121996795ACN 121996795 ACN121996795 ACN 121996795ACN-121996795-A

Abstract

The invention discloses a method and a system for classifying telephone voices in hotel rooms, wherein the method comprises the steps of obtaining the telephone voices in hotel rooms; the method comprises the steps of transferring telephone voices in a hotel to a customer complaint text based on a voice transfer model, sorting the customer complaint text based on a text sorting model, sorting the customer complaint text based on a sentence piece sorting model and a whole sentence sorting model to obtain various customer complaint sorting results, carrying out weighted voting on the various customer complaint sorting results, and carrying out arbitration based on a rule engine to obtain optimized customer complaint sorting results. The technical scheme provided by the invention realizes the classification of the complaints with high accuracy and high recall rate through multi-level and multi-angle model joint processing, optimizes the accuracy of distinguishing the complaints through the cooperation of the sentence-level and whole sentence-level models, realizes multi-level classification of the complaints, expands the application scene of classification of the complaints, meets the complex service requirements, obviously improves the classification performance compared with a single model, and can improve the generalization capability of the model.

Inventors

  • CHEN XUWEI

Assignees

  • 上海华客信息科技有限公司

Dates

Publication Date
20260508
Application Date
20251217

Claims (10)

  1. 1. A method for classifying telephone voices in hotel rooms, comprising the steps of: Acquiring telephone voice in hotel room; Based on a voice transcription model, transcribing telephone voice in the hotel into a customer complaint text; based on the text clause model, the complaint text is subjected to clause; Classifying the complaint texts based on the sentence piece classification model and the whole sentence piece classification model to obtain various complaint classification results; And carrying out weighted voting on various customer complaint classification results, and carrying out arbitration based on a rule engine to obtain an optimized customer complaint classification result.
  2. 2. The method of claim 1, wherein the speech transcription model comprises LSTM, deepSpeech or Conformer.
  3. 3. The method of claim 1, wherein the text sentence model comprises CRF, biLSTM-CRF, or Bert.
  4. 4. The method of claim 1, wherein the text clause model comprises: the input layer is used for carrying out bi-directional context modeling on each character in the input text based on the Bert model, and generating character-level context embedding; the middle layer is used for processing character-level context embedding based on lattice structure flattening and self-attention mechanism of the Flat-Lattice Transformer model to generate enhanced embedding of fusion vocabulary information; And the output layer is used for generating a text clause result based on the CRF model according to the character-level context embedding and the enhanced embedding of the fused vocabulary information.
  5. 5. The method for classifying telephone voices in hotel room of claim 1, the method is characterized in that the method for classifying the telephone voices in the hotel room further comprises the following steps: And (3) performing word segmentation processing on the complaint text by using a word segmentation device, and decomposing the complaint text into sub words.
  6. 6. The method of claim 5, wherein the word segmentation unit comprises BiLSTM-CRF, bert, or Tokenizer.
  7. 7. The hotel in-check telephone speech classification method of claim 1, wherein the sentence slice-level classification model comprises: The input processing layer is used for processing the input text based on a pretrained Bert model to generate a hidden state sequence rich in context information; and the classification layer generates a classification result based on the Softmax classifier according to the hidden state sequence rich in the context information.
  8. 8. The method for classifying telephone voices in hotel room of claim 1, the method is characterized in that the method for classifying the telephone voices in the hotel room further comprises the following steps: Constructing a knowledge base, slicing the history document, encoding the history document into vectors, and storing the vectors into the knowledge base; Vectorizing the complaint text, retrieving the most relevant definition, rule and example from the knowledge base, constructing the prompt word, and performing the prompt design.
  9. 9. The method of claim 1, wherein said performing weighted voting on a plurality of customer complaint classification results, arbitrating based on a rules engine, and obtaining an optimized customer complaint classification result comprises: obtaining various sentence piece classification results and corresponding confidence levels, and obtaining a first customer complaint classification result and corresponding confidence levels through aggregation or weighting treatment; Acquiring a plurality of whole sentence classification results and corresponding confidence levels, and performing aggregation or weighting treatment to acquire a second customer complaint classification result and corresponding confidence levels; And according to the first customer complaint classification and the corresponding confidence level, the second customer complaint classification and the corresponding confidence level, carrying out arbitration based on a rule engine to obtain an optimized customer complaint classification result.
  10. 10. A hotel in-check telephone voice classification system, characterized by comprising, based on the hotel in-check telephone voice classification method of any one of claims 1 to 9: The acquisition module is used for acquiring telephone voices in the hotel room; the transfer module is used for transferring the telephone voice in the hotel to the customer complaint text based on the voice transfer model; The sentence dividing module is used for dividing the complaint text based on the text sentence dividing model; The classification module is used for classifying the complaint texts based on the sentence piece classification model and the whole sentence piece classification model to obtain various complaint classification results; And the arbitration module is used for carrying out weighted voting on various customer complaint classification results, arbitrating based on the rule engine and obtaining an optimized customer complaint classification result.

Description

Method and system for classifying telephone voices in hotel Technical Field The invention relates to the technical field of hotel guest complaint processing, in particular to a method and a system for classifying telephone voices in hotel check-in. Background With the development of hotel services, the processing efficiency and quality of telephone voice feedback in a customer's residence directly affects customer satisfaction and loyalty. In the hotel operation management process, aiming at massive customer communication data, the traditional customer complaint classification method mainly relies on manual processing, and the classification mode has the remarkable defects that (1) the manual processing efficiency is low, the real-time requirements of peak time periods (such as holidays and large-scale conferences) are particularly difficult to meet, the average response time is long, (2) the subjectivity of manual judgment is high, the classification standard is non-uniform, the probability of classifying the same customer complaints into different categories is high, and (3) the labor cost is high, the total expenditure proportion of customer complaint management is large, the 24-hour service coverage is difficult to carry out, and the operation bottleneck exists. In the prior art, although a voice automatic classification method based on a single model exists, the defects of (1) insufficient understanding capability of complex contexts and low classification accuracy, (2) missing of semantic ambiguity processing mechanisms, which cause misclassification of complaints with different synonyms, (3) weak multi-label classification capability, incapability of accurately identifying composite feedback comprising a plurality of complaint points, and (4) limited generalization capability of the model and unsatisfactory recall rate are often existed. These technical defects directly lead to a large number of customer complaints being misclassified, and further require secondary treatment, resulting in increased management costs and prolonged service periods. Disclosure of Invention In view of the defects of the prior art, the invention provides a voice classification method for a hotel in which the classification of customer complaints with high accuracy and high recall rate is realized through multi-level and multi-angle model joint processing, and the complex business requirements are met. In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme: a telephone voice classification method in hotel, comprising the following steps: Acquiring telephone voice in hotel room; Based on a voice transcription model, transcribing telephone voice in the hotel into a customer complaint text; based on the text clause model, the complaint text is subjected to clause; Classifying the complaint texts based on the sentence piece classification model and the whole sentence piece classification model to obtain various complaint classification results; And carrying out weighted voting on various customer complaint classification results, and carrying out arbitration based on a rule engine to obtain an optimized customer complaint classification result. According to one aspect of the invention, the speech transcription model includes LSTM, deepSpeech or Conformer. According to one aspect of the invention, the text clause model includes CRF, biLSTM-CRF, or Bert. According to one aspect of the invention, the text clause model includes: the input layer is used for carrying out bi-directional context modeling on each character in the input text based on the Bert model, and generating character-level context embedding; the middle layer is used for processing character-level context embedding based on lattice structure flattening and self-attention mechanism of the Flat-Lattice Transformer model to generate enhanced embedding of fusion vocabulary information; And the output layer is used for generating a text clause result based on the CRF model according to the character-level context embedding and the enhanced embedding of the fused vocabulary information. According to one aspect of the invention, the in-hotel telephone voice classification method further comprises: And (3) performing word segmentation processing on the complaint text by using a word segmentation device, and decomposing the complaint text into sub words. According to one aspect of the invention, the word segmenter includes BiLSTM-CRF, bert, or Tokenizer. According to one aspect of the invention, the sentence chip classification model comprises: The input processing layer is used for processing the input text based on a pretrained Bert model to generate a hidden state sequence rich in context information; and the classification layer generates a classification result based on the Softmax classifier according to the hidden state sequence rich in the context information. According to one aspect of the invention, the