Method and device for intention recognition

By vector encoding and feature extraction of the call records and logistics order information of existing logistics customers, and using the intention identification model to identify cooperative intentions, the subjectivity and deviation problems of intention recognition in the existing technology are solved, and efficient and accurate intentions are achieved Identify.

CN119005910APending Publication Date: 2024-11-22BEIJING JINGDONG YUANSHENG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411161804.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-22
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the prior art, the identification of cooperation intentions of existing logistics customers depends on text table records, which has subjectivity and deviations, which affects the accuracy of intention identification.

Method used

By obtaining call records and logistics order information between users, vector encoding and feature extraction are performed, using the intention identification model to identify cooperation intentions, reducing manual recording, and improving data objectivity and comprehensiveness.

Benefits of technology

It realizes efficient and accurate intention recognition, reduces workload, avoids interference from subjective factors, and improves the accuracy of intention recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005910A_ABST
    Figure CN119005910A_ABST
Patent Text Reader

Abstract

The invention discloses an intention recognition method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: acquiring a call record between a first user and a second user, wherein communication contents of the first user and the second user in the aspect of business cooperation are recorded in the call record; the logistics order issuing information of the second user is called, and the logistics order issuing information and the user attribute information of the second user are combined into behavior information; and carrying out vector coding on the call record and the behavior information to obtain a call vector and a behavior vector, splicing the call vector and the behavior vector, carrying out feature extraction on the spliced vectors, and identifying the cooperation intention of the second user. According to the embodiment, efficient and accurate intention recognition is realized, a communication table does not need to be manually recorded and filled, the workload is reduced, the data information on which the intention recognition is based is more comprehensive and objective, the interference of subjective factors is avoided, and the intention recognition accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for intention recognition. Background Art

[0002] In the customer management scenario, identifying in advance whether existing customers have the intention to continue cooperation plays an important role in effectively maintaining existing customers. Specifically for the logistics business, the identification of the cooperation intention of existing logistics existing customers is usually done by relevant business personnel calling existing customers and recording the content of the visit in the form of a text table to further evaluate the cooperation intention of existing customers.

[0003] In the process of implementing the present invention, the inventors found that the prior art has the following problems:

[0004] Although the recording method of text tables is convenient for analysis, it is somewhat subjective. Especially when there are vague expressions in the communication content of customers (existing customers), the content recorded in the table can easily deviate from the customer's true intentions, affecting the accuracy of intent identification. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method and device for intent recognition, which realizes efficient and accurate intent recognition. There is no need to manually record and fill out communication forms. After the telephone communication is completed, the data information of this time is collected and recorded, which reduces the workload. Moreover, the data information based on intent recognition is more comprehensive and objective, avoiding interference from subjective factors, and improving the accuracy of intent recognition.

[0006] To achieve the above object, according to one aspect of an embodiment of the present invention, a method for intent recognition is provided, comprising:

[0007] Acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation;

[0008] Retrieving the logistics order information of the second user, and combining the logistics order information and the user attribute information of the second user into behavior information;

[0009] The call record and the behavior information are respectively vector-encoded to obtain a call vector and a behavior vector, the call vector and the behavior vector are concatenated, and features are extracted from the concatenated vectors to identify the cooperation intention of the second user.

[0010] Optionally, obtaining call records between a first user and a second user includes: obtaining at least one call recording between the first user and the second user within a specified time period; converting each of the call recordings into a corresponding call text; detecting whether each of the call texts contains a keyword in a preset intention keyword library, determining a weight value of each of the call texts based on the detection result and the generation time of the call recording corresponding to each of the call texts, and combining each of the call texts with a weight value into a call record.

[0011] Optionally, it is detected whether each of the call texts contains keywords in a preset intention keyword library, and the weight value of each of the call texts is determined according to the detection result and the generation time of the call recording corresponding to each of the call texts, including: for each of the call texts: segmenting the call text to obtain text words, for each text word: performing text similarity matching between the text word and each of the keywords, and according to the matching result, obtaining the keyword corresponding to the text word, and then determining the word weight value of the text word according to the weight value of the keyword; calculating the word weight value of the call text based on the word weight value of each of the text words; querying the generation time of the call recording corresponding to the call text, and obtaining the time weight value of the call text according to a preset time weight allocation rule; and calculating the weight value of the call text according to the word weight value of the call text and the time weight value of the call text.

[0012] Optionally, after converting each of the call recordings into corresponding call texts, the method also includes: identifying whether the fonts in each of the call texts are standardized, and converting non-standard fonts; identifying whether the text in each of the call texts is correct, and correcting incorrect text; and performing text cleaning on each of the call texts to delete stop words and punctuation marks.

[0013] Optionally, each of the call recordings is converted into a corresponding call text, including: for each of the call recordings, using a pre-built large language model to segment the call recording at a time granularity; converting each segmented sub-recording into a spectrogram, and then encoding and decoding the spectrogram to map the sub-text corresponding to each sub-recording; and sequentially splicing each of the sub-texts into a call record according to the corresponding segmentation time.

[0014] Optionally, the call record and the behavior information are respectively vector encoded to obtain a call vector and a behavior vector, the call vector and the behavior vector are spliced, and then the spliced ​​vector is subjected to feature extraction to identify the cooperation intention of the second user, including: inputting the call record and the behavior information into a pre-constructed intention recognition model, and performing word encoding and high-dimensional mapping on each call text in the call record through the call coding network in the intention recognition model to obtain a high-dimensional vector; converting the high-dimensional vector through a self-attention mechanism, and calculating the call vector based on the weight value of each call text; through the behavior coding network in the intention recognition model, vector splicing the discrete information and the continuous information in the behavior information to obtain a spliced ​​vector; then respectively extracting the low-order feature vector and the high-order feature vector of the spliced ​​vector, splicing the low-order feature vector and the high-order feature vector to obtain the behavior vector corresponding to the behavior information; splicing the call vector and the behavior vector, extracting features from the spliced ​​vector through the multi-layer perceptron in the intention recognition model, and outputting the cooperation intention probability value of the second user.

[0015] Optionally, after identifying the cooperation intention of the second user, the method further includes: storing the identified cooperation intention of the second user in a database; in response to receiving an intention identification query request sent by the requesting party, determining the target second user corresponding to the requesting party, and querying and identifying the cooperation intention of the target second user from the database; returning the cooperation intention to the requesting party, so that the requesting party evaluates the true cooperation intention of the target second user based on the cooperation intention and the status data of the target second user.

[0016] According to a second aspect of an embodiment of the present invention, there is provided a device for intent recognition, including:

[0017] A call record acquisition module, used to acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation;

[0018] a behavior information acquisition module, used to retrieve the logistics order information of the second user, and combine the logistics order information and the user attribute information of the second user into behavior information;

[0019] The user intention recognition module is used to respectively perform vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splice the call vector and the behavior vector, perform feature extraction on the spliced ​​vector, and identify the cooperation intention of the second user.

[0020] According to a third aspect of an embodiment of the present invention, there is provided an electronic device for intention recognition, including:

[0021] one or more processors;

[0022] a storage device for storing one or more programs,

[0023] When the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the first aspect of the embodiment of the present invention.

[0024] According to a fourth aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method provided by the first aspect of the embodiment of the present invention is implemented.

[0025] An embodiment of the invention has the following advantages or beneficial effects: by obtaining the call record between the first user and the second user, the call record records the communication content between the first user and the second user in terms of business cooperation; retrieving the logistics order information of the second user, and combining the logistics order information and the user attribute information of the second user into behavior information; respectively performing vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splicing the call vector and the behavior vector, and performing feature extraction on the spliced ​​vectors, a technical solution for identifying the cooperation intention of the second user has been realized, which realizes an efficient and accurate method for intention recognition, without the need for manual recording and filling out of communication forms. After the telephone communication is completed, the data information of this time is collected and recorded, which reduces the workload, and the data information based on the intention recognition is more comprehensive and objective, avoiding the interference of subjective factors, and improving the accuracy of intention recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0027] Figure 1 is a schematic diagram of the main process of the method for intent recognition according to an embodiment of the present invention;

[0028] Figure 2 is a schematic diagram of a process of determining a behavior vector according to an embodiment of the present invention;

[0029] Figure 3 is a schematic diagram of the intention recognition process of an embodiment of the present invention;

[0030] Figure 4 is a schematic diagram of the overall process of intent recognition in an embodiment of the present invention;

[0031] Figure 5 is a schematic diagram of main modules of an apparatus for intent recognition according to an embodiment of the present invention;

[0032] Figure 6is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0033] Figure 7 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0034] It should be noted that in the technical solution of the present invention, the collection / collection, update, analysis, use, transmission, storage and other aspects of user personal information involved are in compliance with the provisions of relevant laws and regulations, are used for legal and reasonable purposes, are not shared, leaked or sold outside of these legal uses, and are subject to supervision and management by national regulatory authorities. Necessary measures should be taken for user personal information to selectively block the use or access to personal information data to prevent illegal access to such personal information data, ensure that persons who have access to personal information data comply with the provisions of relevant laws and regulations, and ensure the security of user personal information. In addition, once these user personal information data are no longer needed, the risks should be minimized by limiting or even prohibiting data collection and / or deleting data.

[0035] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0036] Although the text table recording method in the existing intent recognition method is convenient for analysis, it is somewhat subjective. Especially when there are vague expressions in the communication content of customers (existing customers), the content recorded in the table is likely to deviate from the customer's true intention, affecting the accuracy of intent recognition and cannot meet practical applications well.

[0037] In order to solve the above problems existing in the prior art, the present invention proposes a method for intention recognition, which uses vector coding and feature extraction methods to identify the cooperation intention of the second user based on the acquired call records of the first user and the second user in terms of business cooperation, combined with the current logistics order information of the second user and the user attribute information of the second user. An efficient and accurate method for intention recognition is implemented, without the need to manually record and fill out communication forms. The data information of this telephone communication is collected and recorded, which reduces the workload, and the data information based on intention recognition is more comprehensive and objective, avoiding the interference of subjective factors, and improving the accuracy of intention recognition.

[0038] In the introduction to the embodiments of the present invention, the terms and their meanings are as follows:

[0039] zhconv library: is a lightweight and efficient Python library dedicated to accurate conversion between simplified and traditional Chinese;

[0040] Transformer: is an encoder-decoder structure, which is widely used in natural language processing tasks such as machine translation, text generation and sentiment analysis;

[0041] Embedding: It is a technique for converting discrete variables into continuous vectors, commonly used for text representation in the field of natural language processing.

[0042] Figure 1 is a schematic diagram of the main process of the method for intent recognition according to an embodiment of the present invention, such as Figure 1 As shown, the method for intent recognition according to the embodiment of the present invention includes the following steps S101 to S103.

[0043] Step S101: Acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user regarding business cooperation.

[0044] Specifically, the maintenance of existing customers is an important branch of customer management. According to research, the cost of developing a new customer is five times higher than the cost of maintaining an existing customer. The customer management department will establish management files for all customers, and when the customer's return visit period arrives or when the customer's feedback is received, the account manager will make a telephone visit to the customer under his / her charge to understand the customer's needs and opinions in the current cooperation process, and whether the customer has the intention to continue the cooperation when the cooperation expires. During the telephone visit, the call system will prompt the customer that the current call content will be recorded so that the customer knows and grasps the content of the communication. It should be noted that the first user in the embodiment of the present invention refers to the business personnel of the customer management department, such as the account manager, and the second user refers to the existing customer who receives the telephone visit. The intention recognition system regularly obtains the call recording between the first user and the corresponding second user, and uses the recording-to-text tool to convert the call recording into text format to obtain the corresponding call record.

[0045] According to one embodiment of the present invention, obtaining call records between a first user and a second user includes: obtaining at least one call recording between the first user and the second user within a specified time period; converting each of the call recordings into a corresponding call text; detecting whether each of the call texts contains a keyword in a preset intention keyword library, determining a weight value of each of the call texts based on the detection result and the generation time of the call recording corresponding to each of the call texts, and combining each of the call texts with a weight value into a call record.

[0046] Specifically, under normal circumstances, when the account manager makes a telephone visit to the customer and the call is connected, the call system issues a voice recording prompt, starts recording, and takes the recording of the call time exceeding the preset communication threshold as a valid recording. In addition, in some cases, there may be two or three telephone visits to the same customer within a period of time, for example, the call is interrupted during the call because the customer has something to deal with temporarily, or the customer does not renew the contract according to the agreed time on the phone. Therefore, within a specified time period, such as within a month, multiple call recordings of the first user and the second user may be obtained. At this time, it is necessary to convert each call recording into a corresponding call text. Considering that there are differences in content and time between multiple call recordings, in order to ensure the scientificity and rationality of call record collection, based on business experience, commonly used keywords are preset to form an intention keyword library, such as "can be renewed", "renewal is no problem", "not planning to renew", etc., and each keyword is assigned an appropriate weight. Detect whether each call text contains keywords in the intention keyword library. If the call text contains "can be renewed" or "renewal is no problem", it means that the customer has the intention to continue cooperation. Correspondingly, the weight of this keyword will be larger. If the call text contains "not intending to renew", it means that the customer has no intention to continue cooperation, and the corresponding weight value will be much smaller. Furthermore, in addition to detecting the content of each call text, the generation time of the call recording corresponding to each call text must also be considered. It can be understood that the call text corresponding to the call recording closest to the current moment is more accurate than the call text corresponding to the previous call recording, that is, it can better reflect the customer's current intention. Therefore, the weight value of each call text is determined based on the generation time of the call recording corresponding to the keyword and the call text to distinguish the value of each call text for intent recognition. Finally, the call texts with weight values ​​are combined to obtain call records that are important for subsequent intent recognition.

[0047] According to one embodiment of the present invention, it is detected whether each of the call texts contains keywords in a preset intention keyword library, and the weight value of each of the call texts is determined according to the detection result and the generation time of the call recording corresponding to each of the call texts, including: for each of the call texts: segmenting the call text to obtain text words, for each text word: performing text similarity matching between the text word and each of the keywords, obtaining the keyword corresponding to the text word according to the matching result, and then determining the word weight value of the text word according to the weight value of the keyword; calculating the word weight value of the call text based on the word weight value of each of the text words; querying the generation time of the call recording corresponding to the call text, and obtaining the time weight value of the call text according to a preset time weight allocation rule; and calculating the weight value of the call text according to the word weight value of the call text and the time weight value of the call text.

[0048] Specifically, for the determination of the weight value of each call text, the call text is segmented according to the grammatical meaning to obtain the text words included therein. For each text word, the text similarity between it and each keyword in the intended keyword library is calculated, and the keyword whose text similarity exceeds the preset threshold is selected as the keyword corresponding to the text word. If there are multiple keywords exceeding the threshold, the keyword with the largest text similarity is selected, and then the weight value of the keyword is assigned to the text word to obtain the word weight value of the text word. After traversing all text words, the word weight value of each text word is obtained. It can be understood that there may be a situation where the text word does not match the corresponding keyword. The embodiment of the present invention adds the matched word weight value to the calculation of the word weight value of the call text, and the word weight value that is not matched can be regarded as a non-word weight value. The specific calculation of the word weight value of the call text can calculate the arithmetic mean of the weight values ​​of each matched keyword, or other aggregation algorithms can be used.

[0049] Furthermore, the time element of the call recording is also very important. In the embodiments of the present invention, a time weight allocation rule is preset in advance. For example, taking the call records of one month as an example, the call recordings that occur in the first 10 days have the first time weight, those that occur in the middle 10 days have the second time weight, and those that occur in the remaining days at the end have the third time weight. Correspondingly, the first time weight < the second time weight < the third time weight. Of course, the time weight allocation rule can also be preset with a finer granularity, assigning a corresponding time weight to each day. The embodiments of the present invention do not make specific limitations on the time weight allocation rule. Query the generation time of the call recording corresponding to the call text, and determine the time weight value corresponding to this generation time from the time weight allocation rule to obtain the time weight value of the call text. Finally, calculate the arithmetic mean of the word weight value and the time weight value of the call text to obtain the weight value of the call text. Of course, different proportions can also be assigned to the word weight value and the time weight value of the call text according to needs. For example, the word weight value accounts for 40% and the time weight value accounts for 60%. According to the assigned proportions, combined with the specific word weight value and time weight value of the call text, calculate the actual numerical values of the word weight value and the time weight value in the entire weight value, and then add them up to obtain the weight value of the call text.

[0050] According to another embodiment of the present invention, after converting each of the call recordings into corresponding call texts, the method further includes: identifying whether the fonts in each of the call texts are standard, and performing font conversion on non-standard fonts; identifying whether the words in each of the call texts are correct, and correcting incorrect words; performing text cleaning on each of the call texts to delete stop words and punctuation marks.

[0051] Specifically, in order to ensure the quality of the call records, after converting each call recording into a corresponding call text, it is also necessary to check the call text, identify whether there are traditional Chinese characters in the fonts of each call text. If traditional Chinese characters are found, use a dedicated traditional-simplified converter, such as the zhconv library, although it is not limited to this, to convert the traditional Chinese characters into simplified Chinese characters. It is also possible to use a pre-specified general error-prone character replacement rule to identify whether there are typos in each call text and correct the found typos. In addition to the above font and typo checks, data cleaning can also be performed on the call text. Using the constructed stop word list, delete stop words and punctuation marks such as "ba", "de", "a", etc. in the call text.

[0052] According to another embodiment of the present invention, each of the call recordings is converted into a corresponding call text, including: for each of the call recordings, using a pre-built language macro model to segment the call recording into time granularity; converting each segmented sub-recording into a spectrogram, then encoding and decoding the spectrogram, and mapping to obtain a sub-text corresponding to each of the sub-recordings; and sequentially splicing each of the sub-texts into a call record according to the corresponding segmentation time.

[0053] Specifically, the embodiment of the present invention uses an advanced language large model whisper model to convert call recordings into corresponding call texts. Whisper is a general open source speech recognition model. The whisper model is based on the transformer architecture and uses 680,000 hours of labeled audio data for model training. The model contains 1.55 billion parameters and has good versatility and robustness. It not only supports the conversion of recording files with various accents, but also has good scene generalization capabilities. It can fully adapt to logistics business scenarios, which not only improves the accuracy of conversion, but also ensures conversion efficiency. The whisper model first divides each call recording into time granularity. The model can adaptively adjust the specific time granularity based on the content of the call recording and fully consider the context of the call recording, and can also fix the time granularity. In the business scenario of the embodiment of the present invention, the fixed time granularity meets the use requirements. The use of fixed time granularity can save computing resources and conversion efficiency. A fixed time granularity of 30 seconds is set, and each call recording is divided into 30-second sub-recordings. The waveform fragments of each sub-recording in a call recording are converted into a Mel spectrogram, and the 30-second data is converted into 3000 data point vectors. The data point vectors are then encoded and decoded, and the sub-texts corresponding to each sub-recording are mapped, and the segmentation time is marked on the corresponding sub-text. Of course, the key-value pair structure can also be used for storage, with the segmentation time as the key and the corresponding sub-text as the value. Finally, the corresponding sub-texts are sequentially spliced ​​together according to the segmentation time to obtain the call record.

[0054] Step S102: retrieve the logistics order information of the second user, and combine the logistics order information and the user attribute information of the second user into behavior information.

[0055] Specifically, although accurate and high-quality call records are obtained as described above, the intention expressed by the customer on the phone may be to perfunctorily answer the answer of the first user's account manager. Therefore, in order to further ensure the authenticity and accuracy of the data information, the embodiment of the present invention introduces the logistics order information such as the number of logistics orders and the category of items issued by the second user's customer in recent times, and combines the second user's customer portrait, business categories, business data, customer complaints, claims records and other user attribute information that has a great impact on customer retention, and combines the logistics order information and the user attribute information of the second user (customer) into behavioral information, which is combined with the above-mentioned call records for subsequent intent identification.

[0056] Step S103: respectively perform vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, concatenate the call vector and the behavior vector, perform feature extraction on the concatenated vectors, and identify the cooperation intention of the second user.

[0057] Specifically, based on the acquired call records and the behavior information of the second user, the feature extraction method of principal component analysis is used to firstly vector encode the call records and the behavior information respectively, and then concatenate the two vectors, perform orthogonal transformation on the concatenated vectors to obtain a multi-dimensional vector, and then perform dimensionality reduction processing on the multi-dimensional vectors so that they can be converted into a low-dimensional system with a higher accuracy, thereby identifying whether the second user has the intention to continue cooperation. Of course, a long short-term memory network can also be used to extract features from the concatenated vectors, and the embodiment of the present invention does not specifically limit the feature extraction method.

[0058] According to one embodiment of the present invention, the call record and the behavior information are respectively vector-encoded to obtain a call vector and a behavior vector, the call vector and the behavior vector are concatenated, and then the concatenated vector is subjected to feature extraction to identify the cooperation intention of the second user, including: inputting the call record and the behavior information into a pre-constructed intention recognition model, and performing word encoding and high-dimensional mapping on each call text in the call record through a call coding network in the intention recognition model to obtain a high-dimensional vector; converting the high-dimensional vector through a self-attention mechanism, and calculating the call vector based on the weight value of each call text; concatenating the discrete information and the continuous information in the behavior information through the behavior coding network in the intention recognition model to obtain a concatenated vector; and then extracting the low-order feature vector and the high-order feature vector of the concatenated vector respectively, and concatenating the low-order feature vector and the high-order feature vector to obtain the behavior vector corresponding to the behavior information; concatenating the call vector and the behavior vector, and extracting features from the concatenated vector through a multi-layer perceptron in the intention recognition model, and outputting the cooperation intention probability value of the second user.

[0059] Specifically, in order to improve the recognition accuracy of cooperation intention, the embodiment of the present invention adopts the current mainstream intelligent model prediction method to build an intention recognition model suitable for this business scenario in advance. The intention recognition model mainly includes a call coding network, a behavior coding network and a multi-layer perceptron. First, the call coding network is used to segment the call records, build a vocabulary, assign corresponding vectors to each segmented word, and use embedding to map the vectors into three semantic vectors Q, K, and V in different spaces in high dimensions; considering that there may be only a few words in the call text that contain the customer's specific intentions, the self-attention mechanism is used to calculate the conversion of high-dimensional vectors, which can be calculated specifically by the following formula: in is the number of columns of the Q and K matrices, i.e., the vector dimension. Average pooling is performed based on the converted vectors, and then combined with the weight values ​​of the call texts determined above, the weighted average of the vectors after average pooling of each call text is calculated to obtain the call vector corresponding to the call record.

[0060] Furthermore, for the behavior information of the second user, since it includes discrete information such as the customer's industry information, level information, user portrait, and continuous information such as recent logistics order information, number of claims, and claim amount, it is necessary to first embed the discrete information and perform norm normalization on the continuous information to obtain their respective vector representations, and then concatenate the vectors of the discrete information and the vectors of the continuous information to obtain a concatenated vector. The low-order feature vectors of the concatenated vector are extracted using a three-layer cross network, and the high-order feature vectors of the concatenated vector are extracted using a three-layer deep neural network. Finally, the low-order feature vectors and high-order feature vectors are concatenated to obtain the behavior vector corresponding to the behavior information.

[0061] Furthermore, the call vector and behavior vector obtained above are spliced, and the spliced ​​result is connected to the multi-layer perceptron in the intention recognition model. The multi-layer perceptron extracts features from the spliced ​​vector, and finally the probability value of the cooperation intention of the second user is obtained through the activation function and output.

[0062] Figure 2 This is a schematic diagram of the process of determining the behavior vector of an embodiment of the present invention. Through the behavior encoding network in the intention recognition model, the discrete information and continuous information in the behavior information are vectorized; then a three-layer cross network is used to extract low-order feature vectors, and a three-layer deep network is used to extract high-order feature vectors; finally, the features are spliced ​​to obtain the corresponding behavior vector.

[0063] Figure 3The diagram is a schematic diagram of the intent recognition process of an embodiment of the present invention. The call vector and the behavior vector are concatenated. The call vector in the embodiment of the present invention is 512-dimensional and the behavior vector is 256-dimensional. After the concatenated vector is subjected to layer-by-layer feature extraction by a three-layer perceptron, the dimensions are reduced to 256 and 128 respectively. Finally, the probability of the second user's cooperation intention is obtained through the sigmoid activation function.

[0064] It can be understood that before inputting the call records and behavior information into the pre-built intent recognition model, the method also includes: determining multiple sampled second users from the second user group, collecting historical logistics order information, historical call records, and user attribute information (operating data, user portraits, user complaints, claims records, etc.) of each sampled second user, determining positive samples and negative samples according to the cooperation records of each sampled second user, and dividing and constructing training sets, validation sets, and test sets according to a preset ratio, such as 7:1:2; using the training set, training the preset intent recognition initial model through the cross entropy loss function and gradient descent, updating the parameters of the intent recognition initial model, and then using the test set and validation set to evaluate the intent recognition model after the parameter update, and determine the intent recognition model that has the ability to accurately predict the user's cooperation intention. The specific model evaluation indicators are as follows:

[0065] Precision = predicted positive samples and actually positive samples / all data predicted as positive samples,

[0066] Recall rate = data predicted to be positive samples and actually positive samples / all positive sample data,

[0067] F1 value = 2*precision*recall / (recall+precision).

[0068] According to another embodiment of the present invention, after identifying the cooperation intention of the second user, the method further includes: storing the identified cooperation intention of the second user in a database; in response to receiving an intention identification query request sent by the requesting party, determining the target second user corresponding to the requesting party, and querying and identifying the cooperation intention of the target second user from the database; returning the cooperation intention to the requesting party, so that the requesting party evaluates the true cooperation intention of the target second user based on the cooperation intention and the status data of the target second user.

[0069] Specifically, intent recognition is mainly for the first user account manager. After obtaining the second user's cooperation intention as mentioned above, the intent recognition system will store the second user's cooperation intention in a database, such as the Elasticsearch database, but it is not limited to this. It will also extract the second user's recent logistics order volume, user portrait, complaint record, number of telephone visits, user level and other behavioral information, and associate the extracted information with the call record to form the second user's status data and store it in the database. The account manager can log in to the intent recognition system when he or she first gets to work every day, query all the target second users he or she is responsible for through the query interface, and obtain the cooperation intention of the target second user that the intent recognition system has already identified in advance, as well as the status data of the associated target second users. Based on the query results, the account manager comprehensively evaluates the target second user's real cooperation intention, so as to accurately locate potential lost customers, and formulate corresponding recovery strategies for customers with loss risks in combination with the historical behavior in the user's status data to reduce customer loss.

[0070] Figure 4 It is a schematic diagram of the overall process of intent recognition of an embodiment of the present invention. The intent recognition system retrieves the call recordings of the first user and the second user, as well as the logistics order information of the second user. The logistics order information and the user attribute information of the second user (business data, specific behaviors such as complaints and claims, and information such as user portraits) are combined into behavior information, and the call recordings are converted into corresponding call records using the whisper language large model. The feature engineering of the call records and behavior information is started separately. The call records and behavior information are first vector-encoded to obtain call vectors and behavior vectors, and the call vectors and behavior vectors are spliced. The spliced ​​vectors are input into the trained intent recognition model for feature extraction to obtain the cooperation intention of the second user. Finally, the obtained cooperation intention and the status data of the second user composed of the call records and the user's behavior information are stored in the database, and data structured processing is performed so that when an intent recognition query request is received from the front end, the identified cooperation intention and the corresponding status data are displayed to the front end for reference by the customer manager.

[0071] The embodiment of the present invention also experimentally verified the method of intent recognition, and the verification results are shown in Table 1 below. Through the propensity score matching algorithm, the differences between customers are eliminated, and the customer data is diverted to divide the customers into an experimental group and a control group. The experimental group identifies the cooperation intention of the second user based on the intention recognition method of the embodiment of the present invention, and the control group is determined by the first user's account manager based on the call recording. Finally, the number of 200 customers in the experimental group who accurately identified the customer intention was 133, with an accuracy rate of 66.5%; the number of 200 customers in the control group who accurately identified the customer intention was 102, with an accuracy rate of 51%. The experimental group was improved by 30.3% compared with the control group.

[0072] The embodiment of the present invention mainly identifies the intention of existing customers to continue to cooperate. In order to enable account managers to accurately estimate which customers have strong retention intentions and which customers have potential loss risks, and to prevent account managers from misjudging intentions due to perfunctory answers to questions, call records and customer behavior information are combined as basic data for intent identification, and accurate conversion is performed based on call recordings to generate high-quality call records. Automatic, accurate, and efficient intent identification by the intention recognition model can greatly reduce the work efficiency and recognition accuracy of account managers, and is more helpful for account managers to formulate accurate recovery strategies to minimize the loss of existing customers.

[0073] Figure 5 Schematic diagram of the main modules of the apparatus for intent recognition according to an embodiment of the present invention. Figure 5 As shown, the device 500 for intention recognition mainly includes a call record acquisition module 501 , a behavior information acquisition module 502 and a user intention recognition module 503 .

[0074] A call record acquisition module 501 is used to acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation;

[0075] The behavior information acquisition module 502 is used to retrieve the logistics order information of the second user, and combine the logistics order information and the user attribute information of the second user into behavior information;

[0076] The user intention identification module 503 is used to respectively perform vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splice the call vector and the behavior vector, perform feature extraction on the spliced ​​vector, and identify the cooperation intention of the second user.

[0077] According to one embodiment of the present invention, the call record acquisition module 501 is also used to: obtain at least one call recording between the first user and the second user within a specified time period; convert each of the call recordings into a corresponding call text; detect whether each of the call texts contains a keyword in a preset intention keyword library, determine the weight value of each of the call texts based on the detection result and the generation time of the call recording corresponding to each of the call texts, and combine each of the call texts with a weight value into a call record.

[0078] According to another embodiment of the present invention, the call record acquisition module 501 is also used for: for each of the call texts: segmenting the call text to obtain text words, for each text word: performing text similarity matching between the text word and each of the keywords, obtaining the keyword corresponding to the text word according to the matching result, and then determining the word weight value of the text word according to the weight value of the keyword; calculating the word weight value of the call text based on the word weight value of each of the text words; querying the generation time of the call recording corresponding to the call text, and obtaining the time weight value of the call text according to a preset time weight allocation rule; and calculating the weight value of the call text according to the word weight value of the call text and the time weight value of the call text.

[0079] According to another embodiment of the present invention, the device 500 for intention recognition also includes an optimization module (not shown in the figure), which is used to: after converting each of the call recordings into a corresponding call text, identify whether the fonts in each of the call texts are standardized, and perform font conversion on non-standard fonts; identify whether the text in each of the call texts is correct, and correct incorrect text; perform text cleaning on each of the call texts, and delete stop words and punctuation marks.

[0080] According to another embodiment of the present invention, the call record acquisition module 501 is further used to: for each of the call recordings, use a pre-built language large model to segment the call recordings into time granularity; convert each segmented sub-recording into a spectrogram, and then encode and decode the spectrogram to map and obtain a sub-text corresponding to each of the sub-recordings; and sequentially splice each of the sub-texts into a call record according to the corresponding segmentation time.

[0081] According to another embodiment of the present invention, the user intention recognition module 503 is also used to: input the call record and the behavior information into a pre-constructed intention recognition model, and perform word encoding and high-dimensional mapping on each call text in the call record through the call coding network in the intention recognition model to obtain a high-dimensional vector; convert the high-dimensional vector through a self-attention mechanism, and calculate the call vector based on the weight value of each call text; through the behavior coding network in the intention recognition model, vector splice the discrete information and continuous information in the behavior information to obtain a spliced ​​vector; then respectively extract the low-order feature vector and the high-order feature vector of the spliced ​​vector, splice the low-order feature vector and the high-order feature vector to obtain the behavior vector corresponding to the behavior information; splice the call vector and the behavior vector, perform feature extraction on the spliced ​​vector through the multi-layer perceptron in the intention recognition model, and output the cooperation intention probability value of the second user.

[0082] According to another embodiment of the present invention, the intention identification device 500 also includes an intention determination module (not shown in the figure), which is used to: after identifying the cooperation intention of the second user, store the identified cooperation intention of the second user in a database; in response to receiving an intention identification query request sent by the requesting party, determine the target second user corresponding to the requesting party, and query and identify the cooperation intention of the target second user from the database; return the cooperation intention to the requesting party, so that the requesting party can evaluate the true cooperation intention of the target second user based on the cooperation intention and the status data of the target second user.

[0083] Figure 6 is an exemplary system architecture diagram to which embodiments of the present invention can be applied.

[0084] like Figure 6 As shown, system architecture 600 may include terminal devices 601, 602, 603, network 604 and server 605. Network 604 is used to provide a medium for communication links between terminal devices 601, 602, 603 and server 605. Network 604 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0085] Users can use terminal devices 601, 602, 603 to interact with server 605 through network 604 to receive or send messages, etc. Various communication client applications, such as intent recognition applications, etc., can be installed on terminal devices 601, 602, 603 (only as an example).

[0086] The terminal devices 601 , 602 , and 603 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers, etc.

[0087] Server 605 may be a server that provides various services, such as a background management server that provides support for the intention recognition performed by users using terminal devices 601, 602, and 603 (for example only). The background management server may obtain the call record between the first user and the second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation; retrieve the logistics order information of the second user, and combine the logistics order information and the user attribute information of the second user into behavior information; respectively perform vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, concatenate the call vector and the behavior vector, perform feature extraction on the concatenated vector, identify the cooperation intention of the second user, and feed back the processing result (such as cooperation intention, etc. - for example only) to the terminal device.

[0088] It should be noted that the method for intent recognition provided in the embodiment of the present invention is generally executed by the server 605 , and accordingly, the device for intent recognition is generally set in the server 605 .

[0089] It should be understood that Figure 6 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0090] Reference below Figure 7 , which shows a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. Figure 7 The terminal device or server shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0091] like Figure 7 As shown, the computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the system 700 are also stored. The CPU 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0092] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.

[0093] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the central processing unit (CPU) 701, the above-mentioned functions defined in the system of the present invention are executed.

[0094] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination thereof.

[0095] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0096] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be set in a processor, for example, it may be described as: a processor includes: a call record acquisition module, a behavior information acquisition module and a user intention recognition module.

[0097] The names of these modules do not, in some cases, limit the modules themselves. For example, a call record acquisition module may also be described as a “module for acquiring call records between a first user and a second user”.

[0098] On the other hand, the present invention also provides a computer-readable medium, which may be included in the device described in the embodiment; or it may exist independently without being assembled into the device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by a device, the device includes: obtaining a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation; retrieving the logistics order information of the second user, and combining the logistics order information and the user attribute information of the second user into behavior information; respectively performing vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splicing the call vector and the behavior vector, performing feature extraction on the spliced ​​vector, and identifying the cooperation intention of the second user.

[0099] The technical solution according to the embodiment of the present invention has the following advantages or beneficial effects: by obtaining the call record between the first user and the second user, the call record records the communication content between the first user and the second user in terms of business cooperation; retrieving the logistics order information of the second user, and combining the logistics order information and the user attribute information of the second user into behavior information; respectively performing vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splicing the call vector and the behavior vector, performing feature extraction on the spliced ​​vector, and identifying the cooperation intention of the second user. The technical solution realizes an efficient and accurate method for intention recognition, without the need for manual recording and filling out of communication forms. After the telephone communication is completed, the data information of this time is collected and recorded, which reduces the workload, and the data information based on the intention recognition is more comprehensive and objective, avoiding the interference of subjective factors, and improving the accuracy of intention recognition.

[0100] The specific implementation methods described herein do not constitute limitations on the scope of protection of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A method for intent recognition, characterized in that: include: Acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation; Retrieving the logistics order information of the second user, and combining the logistics order information and the user attribute information of the second user into behavior information; The call record and the behavior information are respectively vector-encoded to obtain a call vector and a behavior vector, the call vector and the behavior vector are concatenated, and features are extracted from the concatenated vectors to identify the cooperation intention of the second user.

2. The method according to claim 1, characterized in that Get the call records between the first user and the second user, including: Obtain at least one call recording between the first user and the second user within a specified time period; Convert each of the call recordings into corresponding call texts; Detect whether each of the call texts contains keywords in a preset intention keyword library, determine the weight value of each of the call texts based on the detection result and the generation time of the call recording corresponding to each of the call texts, and combine the call texts with weight values ​​into a call record.

3. The method according to claim 2, characterized in that Detecting whether each of the call texts contains a keyword in a preset intention keyword library, and determining a weight value of each of the call texts according to the detection result and the generation time of the call recording corresponding to each of the call texts, including: For each of the said calls text: Segmenting the call text to obtain text words, and for each text word: performing text similarity matching between the text word and each of the keywords, obtaining the keyword corresponding to the text word according to the matching result, and then determining the word weight value of the text word according to the weight value of the keyword; Calculating a word weight value of the call text based on the word weight value of each of the text words; Querying the generation time of the call recording corresponding to the call text, and obtaining the time weight value of the call text according to a preset time weight allocation rule; The weight value of the call text is calculated according to the word weight value of the call text and the time weight value of the call text.

4. The method according to claim 2, characterized in that: After converting each of the call recordings into a corresponding call text, the method further includes: Identifying whether the fonts in each of the call texts are standard, and performing font conversion on non-standard fonts; Identify whether the words in each of the call texts are correct and correct the wrong words; The call texts are cleaned to remove stop words and punctuation marks.

5. The method according to claim 2, characterized in that: Convert each of the call recordings into corresponding call texts, including: For each of the call recordings, a pre-built language model is used to segment the call recordings into time granularities; Converting each segmented sub-recording into a spectrogram, encoding and decoding the spectrogram, and mapping to obtain a sub-text corresponding to each sub-recording; The sub-texts are sequentially spliced ​​into a call record according to the corresponding segmentation time.

6. The method according to claim 2, characterized in that The call record and the behavior information are respectively vector-encoded to obtain a call vector and a behavior vector, the call vector and the behavior vector are concatenated, and then feature extraction is performed on the concatenated vectors to identify the cooperation intention of the second user, including: Inputting the call records and the behavior information into a pre-built intent recognition model, performing word encoding and high-dimensional mapping on each call text in the call records through a call encoding network in the intent recognition model to obtain a high-dimensional vector; converting the high-dimensional vector through a self-attention mechanism, and calculating a call vector based on the weight value of each call text; Through the behavior encoding network in the intention recognition model, discrete information and continuous information in the behavior information are vector-concatenated to obtain a concatenated vector; low-order feature vectors and high-order feature vectors of the concatenated vector are extracted respectively, and the low-order feature vectors and the high-order feature vectors are concatenated to obtain a behavior vector corresponding to the behavior information; The call vector and the behavior vector are concatenated, and features are extracted from the concatenated vectors through a multi-layer perceptron in the intention recognition model to output a probability value of the cooperation intention of the second user.

7. The method according to claim 1, characterized in that After identifying the cooperation intention of the second user, the method further includes: storing the identified cooperation intention of the second user in a database; In response to receiving the intention identification query request sent by the requesting party, determining the target second user corresponding to the requesting party, and querying and identifying the cooperation intention of the target second user from the database; The cooperation intention is returned to the requesting party, so that the requesting party evaluates the real cooperation intention of the target second user according to the cooperation intention and the status data of the target second user.

8. A device for intention recognition, characterized in that: include: A call record acquisition module, used to acquire a call record between a first user and a second user, wherein the call record records the communication content between the first user and the second user in terms of business cooperation; a behavior information acquisition module, used to retrieve the logistics order information of the second user, and combine the logistics order information and the user attribute information of the second user into behavior information; The user intention recognition module is used to respectively perform vector encoding on the call record and the behavior information to obtain a call vector and a behavior vector, splice the call vector and the behavior vector, perform feature extraction on the spliced ​​vector, and identify the cooperation intention of the second user.

9. A mobile electronic device terminal, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.