CN117633619A - Customer service telephone conversation text classification method and device - Google Patents

Abstract

The application discloses a customer service telephone conversation text classification method and device. Wherein the method comprises the following steps: obtaining a text to be classified, wherein the text to be classified is obtained by converting customer service call records; analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; and analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified. The method and the device solve the technical problem of low text classification efficiency of the customer service telephone in the related technology.

Description

Customer service telephone conversation text classification method and device

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a customer service telephone conversation text classification method and device.

Background

The business around the customer service center has a large number of scenes of intensive manpower, high frequency and more data accumulation, so that the work which is finished by relying on a large amount of manpower is needed to pass, for example: in each period, the text naturally expressed in the incoming call of the current user is automatically classified in a summary mode according to the established service dimension, and currently, in the related technology, the summary classification work is often completed manually, so that the efficiency is low and mistakes are easy to occur.

Disclosure of Invention

The embodiment of the application provides a customer service telephone conversation text classification method and device, which are used for at least solving the technical problem of low customer service telephone text classification efficiency in the related technology.

According to one aspect of the embodiments of the present application, there is provided a customer service telephone conversation text classification method, including: obtaining a text to be classified, wherein the text to be classified is obtained by converting customer service call records; analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; and analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

Optionally, analyzing the text to be classified by using a first classification model and a second classification model to obtain a first prediction result and a second prediction result, including: and sequentially classifying the text to be classified by adopting the first classification model to obtain the first prediction result, wherein the first prediction result comprises: the classification result of the text to be classified under three levels, wherein the classification result of each level is a sub-category of the classification result of the last level; dividing the text to be classified into a first text set and a second text set by adopting the second classification model, and determining the second prediction result according to the first text set and the second text set, wherein the second prediction result comprises: and the first text set is formed by splitting the text to be classified into a plurality of clauses, and the second text set is formed by splitting the text to be classified into a plurality of segmentations.

Optionally, determining the second prediction result according to the first text set and the second text set includes: extracting features of the first text set and the second text set by adopting the second classification model to obtain a first vector set and a second vector set, fusing vectors in the first vector set to obtain a first fused vector, and fusing vectors in the second vector set to obtain a second fused vector; and respectively analyzing the first fusion vector and the second fusion vector by adopting the second classification model to obtain the second prediction result.

Optionally, the first classification model is trained by: obtaining a training dataset comprising: a plurality of levels of classification labels; training a plurality of levels of classification tasks in the first classification model by using the training data set until the first classification model converges, wherein the plurality of levels of classification tasks share the same loss function.

Optionally, fusing the first prediction result and the second prediction result to obtain a target prediction result includes: acquiring weights of the first prediction result and the second prediction result; and carrying out weighted summation on the first predicted result and the second predicted result to obtain the target predicted result.

Optionally, analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified, including: determining classification results of a plurality of levels indicated by the target prediction result as a plurality of levels of a preset search tree, taking the classification result of the highest level in the plurality of levels as a root node of the preset search tree, and taking the classification results of other levels in the plurality of levels as child nodes under the root node; sequentially selecting the next level node with highest prediction probability from the root node until the leaf node of the preset search tree is a classification node of the lowest level in the multiple levels; and determining classification results corresponding to all nodes from the root node to the leaf node as the final classification result.

Optionally, selecting a next layer node with highest prediction probability from the root node in turn until a leaf node of the preset search tree includes: determining a plurality of node heuristic cost values in each level of the preset search tree by adopting a preset heuristic function, wherein the heuristic cost values are the prediction probabilities of the plurality of nodes; deleting the nodes with the heuristic cost value smaller than a preset threshold value in each level to obtain a plurality of candidate nodes in each level; and selecting the candidate node with the highest heuristic cost value from the plurality of candidate nodes of each hierarchy in turn as the next-layer node until the leaf node of the preset search tree.

According to another aspect of the embodiments of the present application, there is also provided a customer service telephone conversation text classification device, including: the acquisition module is used for acquiring texts to be classified, wherein the texts to be classified are obtained by converting customer service call records; the prediction module is used for respectively analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; and the classification module is used for analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

According to still another aspect of the embodiments of the present application, there is further provided a non-volatile storage medium, in which a program is stored, where when the program runs, a device on which the non-volatile storage medium is controlled to execute the above method for classifying text of a customer service telephone conversation.

According to still another aspect of the embodiments of the present application, there is also provided a computer apparatus including: the system comprises a memory and a processor, wherein the processor is used for running a program stored in the memory, and the program runs to execute the customer service telephone conversation text classification method.

In the embodiment of the application, a text to be classified is acquired, and the text to be classified is obtained by converting customer service call records; analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; the method comprises the steps of analyzing the classification result indicated by the target prediction result to obtain the final classification result of the text to be classified, classifying the text to be classified through two classification models respectively, fusing the prediction results of the two classification models to obtain the target prediction result, and finally analyzing the target prediction result to obtain the final prediction result, so that the aim of classifying the text to be classified by using artificial intelligence is fulfilled, the technical effect of improving the text classification efficiency is achieved, and the technical problem of low text classification efficiency of customer service telephones in related technologies is solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:

FIG. 1 is a block diagram of the hardware architecture of a computer terminal (or mobile device) for a customer service telephone conversation text classification method according to an embodiment of the present application;

FIG. 2 is a flow diagram of a method of classifying customer service telephone conversation text according to the present application;

FIG. 3 is a flow chart diagram illustration of an alternative data preprocessing according to an embodiment of the present application;

FIG. 4 is a flow diagram of another method of classification of customer service telephone conversation text according to the present application;

FIG. 5 is a schematic diagram of an alternative first classification model classification flow according to the present application;

FIG. 6 is a schematic diagram of an alternative second classification model classification flow according to the present application;

FIG. 7 is a schematic illustration of an alternative final sort result production flow according to the present application;

fig. 8 is a schematic structural diagram of an alternative customer service telephone conversation text classification device according to an embodiment of the present application.

Detailed Description

In order to make the present application solution better understood by those skilled in the art, the following description will be made in detail and with reference to the accompanying drawings in the embodiments of the present application, it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.

It should be noted that the terms "first," "second," and the like in the description and claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.

The method embodiments provided by the embodiments of the present application may be performed in a mobile terminal, a computer terminal, or similar computing device. Fig. 1 shows a block diagram of a hardware architecture of a computer terminal (or mobile device) for implementing a customer service telephone conversation text classification method. As shown in fig. 1, the computer terminal 10 (or mobile device 10) may include one or more processors 102 (shown as 102a, 102b, … …,102 n) which may include, but are not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA, a memory 104 for storing data, and a transmission module 106 for communication functions. In addition, the method may further include: a display, an input/output interface (I/O interface), a Universal Serial BUS (USB) port (which may be included as one of the ports of the BUS), a network interface, a power supply, and/or a camera. It will be appreciated by those of ordinary skill in the art that the configuration shown in fig. 1 is merely illustrative and is not intended to limit the configuration of the electronic device described above. For example, the computer terminal 10 may also include more or fewer components than shown in FIG. 1, or have a different configuration than shown in FIG. 1.

It should be noted that the one or more processors 102 and/or other data processing circuits described above may be referred to generally herein as "data processing circuits. The data processing circuit may be embodied in whole or in part in software, hardware, firmware, or any other combination. Furthermore, the data processing circuitry may be a single stand-alone processing module, or incorporated, in whole or in part, into any of the other elements in the computer terminal 10 (or mobile device). As referred to in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of the path of the variable resistor termination to interface).

The memory 104 may be used to store software programs and modules of application software, such as a program instruction/data storage device corresponding to the classification method of customer service phone conversation text in the embodiment of the present application, and the processor 102 executes the software programs and modules stored in the memory 104, thereby performing various functional applications and data processing, that is, implementing the classification method of customer service phone conversation text as described above. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory located remotely from the processor 102, which may be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission means 106 is arranged to receive or transmit data via a network. The specific examples of the network described above may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module for communicating with the internet wirelessly.

The display may be, for example, a touch screen type Liquid Crystal Display (LCD) that may enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

In the above operating environment, the embodiment of the present application provides a method for classifying text of a customer service telephone conversation, as shown in fig. 2, the method includes the following steps:

step S202, obtaining a text to be classified, wherein the text to be classified is obtained by converting customer service call records;

step S204, analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model;

and S206, analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

Through the steps, the text to be classified can be obtained, and the text to be classified is obtained by converting customer service call records; analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; the method comprises the steps of analyzing the classification result indicated by the target prediction result to obtain the final classification result of the text to be classified, classifying the text to be classified through two classification models respectively, fusing the prediction results of the two classification models to obtain the target prediction result, and finally analyzing the target prediction result to obtain the final prediction result, so that the aim of classifying the text to be classified by using artificial intelligence is fulfilled, the technical effect of improving the text classification efficiency is achieved, and the technical problem of low text classification efficiency of customer service telephones in related technologies is solved.

It should be noted that, the text to be classified may be obtained by preprocessing a customer service call record, and an optional manner is as follows: step 1: data preprocessing, namely converting call records needing to be classified into call texts, and filtering invalid data for all call records needing to be classified, wherein the invalid data at least comprises the following steps: calls that employ conventional manual telephone nodule operations, calls that are abnormally hung up/abnormally muted/marked as invalid, calls that have no corresponding transcribed text or have too low of a transcribed text availability, etc. Step 2: text preprocessing, marking service related hot words, clustering similar data in dialogue texts and extracting the hot words, constructing a key knowledge base, improving the sensitivity of a model to service types, and improving the prediction accuracy of the model; the irrelevant words are marked to form a corresponding characteristic word library, and the corresponding characteristic word library is used for hiding the part of irrelevant information during semantic processing, so that input information purification is realized; and providing training sample corpus, namely, manually judging by an expert, taking 100% correct call text classified by the nodules as a training sample, and inputting the training sample corpus into a model to realize learning driving.

Fig. 3 shows a flow chart of data preprocessing, as shown in fig. 3, the original sample data is input through the telephone nodule classification application program, the classification samples needing training are screened through data cleaning and preprocessing, and the telephone nodule classification model is trained and tested.

Fig. 4 illustrates another customer service telephone conversation text classification method, and as shown in fig. 4, the first classification model may be a BERT multitasking learning model (a base semantic vector bi-directional encoder model) and the second classification model may be a HAN multitasking learning model (an attention handling model for multi-level text).

Steps S202 to S206 are described below by specific embodiments.

In step S204, the text to be classified is analyzed by using the first classification model and the second classification model, and the specific steps for obtaining the first prediction result and the second prediction result are as follows: and sequentially classifying the text to be classified by adopting the first classification model to obtain the first prediction result, wherein the first prediction result comprises: the classification result of the text to be classified under three levels, wherein the classification result of each level is a sub-category of the classification result of the last level; dividing the text to be classified into a first text set and a second text set by adopting the second classification model, and determining the second prediction result according to the first text set and the second text set, wherein the second prediction result comprises: and the first text set is formed by splitting the text to be classified into a plurality of clauses, and the second text set is formed by splitting the text to be classified into a plurality of segmentations.

Wherein the first classification model is trained by: obtaining a training dataset comprising: a plurality of levels of classification labels; training a plurality of levels of classification tasks in the first classification model by using the training data set until the first classification model converges, wherein the plurality of levels of classification tasks share the same loss function.

Specifically, taking a first classification model as a multi-task learning model created based on a BERT model as an example, as shown in fig. 5, the first layer is the BERT model and is responsible for extracting features of texts to be classified, the second layer is a plurality of fully connected layers (for example: three) and is responsible for classifying the texts to be classified, the input of the first classification model is the processed texts to be classified and keywords, and the output of the first classification model is classification results of a plurality of categories, for example: the method comprises the steps of classifying a text to be classified into a business major class, classifying results of sub-products or sub-services under the business major class, classifying results of problems or requirements under the sub-products to which the text to be classified belongs, and understanding that the model utilizes the pretraining capability of BERT, takes text classification tasks of different levels as a plurality of sub-tasks, and realizes knowledge migration and complementation among the multiple tasks through sharing parameters and optimizing targets. The model can effectively improve the accuracy and the robustness of text classification, and simultaneously reduces training time and resource consumption.

It should be noted that, the classification tasks of multiple levels share the same loss function, errors of classification tasks of different levels can be concentrated into the same loss function to calculate to obtain gradients, and the gradients are propagated in opposite directions, so that the model can learn association relations between labels of different levels at the same time in the training process, even if label prediction on a certain level is wrong, but prediction on other levels is wrong, the whole sample still generates errors, so that the model can adjust the errors, and finally the model can converge towards a direction of 'all-level label prediction correct', and different weights can be allocated according to different subtasks when the loss function values of multiple levels are combined.

Wherein the second prediction result may be determined by: extracting features of the first text set and the second text set by adopting the second classification model to obtain a first vector set and a second vector set, fusing vectors in the first vector set to obtain a first fused vector, and fusing vectors in the second vector set to obtain a second fused vector; and respectively analyzing the first fusion vector and the second fusion vector by adopting the second classification model to obtain the second prediction result.

As shown in fig. 6, in the case that the second classification model is a multi-task learning model created based on the HAN model, the text feature extraction model of the first classification model is replaced by the HAN model from the BERT model, the text to be identified is divided into a first text set and a second text set by using the hierarchical structure of the HAN model, coding and attention calculation are performed respectively, then vectors in the first text set are fused into a first fusion vector, vectors in the second text set are fused into a second fusion vector, semantic information and importance of the text to be classified on different levels are captured, and the precision of text classification is improved.

In some embodiments of the present application, the method for fusing the first prediction result and the second prediction result to obtain the target prediction result is as follows: acquiring weights of the first prediction result and the second prediction result; and carrying out weighted summation on the first predicted result and the second predicted result to obtain the target predicted result.

Specifically, the weights of the first classification model and the second classification model are dynamically adjusted according to the performances of the first classification model and the second classification model on the verification set, so that an optimal fusion result is obtained. The method comprises the following steps:

first, the training set is divided into a training subset and a verification subset, the training subset is used to train the first classification model and the second classification model, the verification subset is used to evaluate the accuracy or other index of the first classification model and the second classification model, and the weight is calculated according to the performance. An alternative approach is to use an inverse scaling function to calculate the weights, i.e. the more well behaved model weights, the less poorly behaved model weights.

Finally, the outputs of the first classification model and the second classification model are multiplied by their weights and summed to obtain the final output (target prediction result).

In some embodiments of the present application, the final classification result may be determined by: determining classification results of a plurality of levels indicated by the target prediction result as a plurality of levels of a preset search tree, taking the classification result of the highest level in the plurality of levels as a root node of the preset search tree, and taking the classification results of other levels in the plurality of levels as child nodes under the root node; sequentially selecting the next level node with highest prediction probability from the root node until the leaf node of the preset search tree is a classification node of the lowest level in the multiple levels; and determining classification results corresponding to all nodes from the root node to the leaf node as the final classification result.

The next layer of nodes with highest prediction probability are sequentially selected from the root node until the leaf nodes of the preset search tree are specifically processed as follows: determining a plurality of node heuristic cost values in each level of the preset search tree by adopting a preset heuristic function, wherein the heuristic cost values are the prediction probabilities of the plurality of nodes; deleting the nodes with the heuristic cost value smaller than a preset threshold value in each level to obtain a plurality of candidate nodes in each level; and selecting the candidate node with the highest heuristic cost value from the plurality of candidate nodes of each hierarchy in turn as the next-layer node until the leaf node of the preset search tree.

In an actual application scenario, a heuristic graph Search algorithm may be adopted, where a Search tree is built for using breadth-first policies in a bundle Search (Beam Search) process, as shown in fig. 7, at each layer of the tree, nodes are ordered according to heuristic costs, then only a predetermined number of nodes are left, only those nodes continue to expand at the next layer, and other nodes will be cut. If the width of the bundle is infinite, the bundle search is converted into the width-first search, the search tree is input into a plurality of levels of classification results after model fusion, and as shown in fig. 7, when the bundle search algorithm achieves classification of the text to be classified, the classification results of the plurality of levels are regarded as a plurality of levels of the search tree, the first level classification result is used as a root node for expansion search, and finally, a path with the highest probability is selected as a final prediction result.

In an alternative, the method comprises the steps of:

step 1, defining a heuristic function h (n) that can estimate the cost from the current node n to the target node (i.e. the final classification result), in an alternative way using the predicted probability value of the current node n as the value of the heuristic function h (n).

It can be appreciated that the prediction probability is the probability that the classification result is the class corresponding to the current node.

Step 2, defining a pruning strategy p (n) to determine to keep or discard the current node n, for example: judging whether the probability value of the current node n is larger than or equal to theta through a preset threshold value theta, and if so, reserving; if not, discard.

Finally, a depth-first search is performed starting from the root node and using a heuristic function h (n) and pruning strategy p (n) at each level to select the most promising candidate nodes until the target node or leaf node of the search tree is reached. And finally selecting the path with the highest probability as a final classification result.

As shown in fig. 7, the classification results of the first level are a and C, the second level is B and E, the third level is D and D, and finally two classification results are ABD and CED, respectively.

The embodiment of the application provides a customer service telephone conversation text classification device, as shown in fig. 8, comprising:

the obtaining module 80 is configured to obtain a text to be classified, where the text to be classified is obtained by converting a customer service call record;

the prediction module 82 is configured to analyze the text to be classified by using a first classification model and a second classification model, respectively, to obtain a first prediction result and a second prediction result, and fuse the first prediction result and the second prediction result to obtain a target prediction result, where the target prediction result is used to indicate a classification result of the text to be classified, and a feature extraction mode of the first classification model is different from that of the second classification model;

and the classification module 84 is configured to analyze the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

The prediction module 82 includes: the classification sub-module is used for sequentially classifying the text to be classified in multiple levels by adopting the first classification model to obtain the first prediction result, wherein the first prediction result comprises: the classification result of the text to be classified under three levels, wherein the classification result of each level is a sub-category of the classification result of the last level; dividing the text to be classified into a first text set and a second text set by adopting the second classification model, and determining the second prediction result according to the first text set and the second text set, wherein the second prediction result comprises: and the first text set is formed by splitting the text to be classified into a plurality of clauses, and the second text set is formed by splitting the text to be classified into a plurality of segmentations.

The classifying sub-module comprises: the prediction unit is used for extracting features of the first text set and the second text set by adopting the second classification model to obtain a first vector set and a second vector set, fusing vectors in the first vector set to obtain a first fused vector, and fusing vectors in the second vector set to obtain a second fused vector; and respectively analyzing the first fusion vector and the second fusion vector by adopting the second classification model to obtain the second prediction result.

The prediction module 82 further includes: the training sub-module and the fusion sub-module are used for acquiring a training data set, wherein the training data set comprises: a plurality of levels of classification labels; training a plurality of levels of classification tasks in the first classification model by using the training data set until the first classification model converges, wherein the plurality of levels of classification tasks share the same loss function.

The fusion sub-module is used for acquiring weights of the first prediction result and the second prediction result; and carrying out weighted summation on the first predicted result and the second predicted result to obtain the target predicted result.

The classification module 84 includes: a determining submodule, configured to determine classification results of multiple levels indicated by the target prediction result as multiple levels of a preset search tree, take a classification result of a highest level in the multiple levels as a root node of the preset search tree, and take classification results of other levels in the multiple levels as child nodes under the root node; sequentially selecting the next level node with highest prediction probability from the root node until the leaf node of the preset search tree is a classification node of the lowest level in the multiple levels; and determining classification results corresponding to all nodes from the root node to the leaf node as the final classification result.

The determining sub-module includes: a determining unit, configured to determine heuristic cost values of a plurality of nodes in each level of the preset search tree by using a preset heuristic function, where the heuristic cost values are prediction probabilities of the plurality of nodes; deleting the nodes with the heuristic cost value smaller than a preset threshold value in each level to obtain a plurality of candidate nodes in each level; and selecting the candidate node with the highest heuristic cost value from the plurality of candidate nodes of each hierarchy in turn as the next-layer node until the leaf node of the preset search tree.

According to still another aspect of the embodiments of the present application, there is further provided a non-volatile storage medium, in which a program is stored, where when the program runs, a device on which the non-volatile storage medium is controlled to execute the above method for classifying text of a customer service telephone conversation.

According to still another aspect of the embodiments of the present application, there is also provided a computer apparatus including: the system comprises a memory and a processor, wherein the processor is used for running a program stored in the memory, and the program runs for executing the customer service telephone conversation text classification method, and the method comprises the following steps: obtaining a text to be classified, wherein the text to be classified is obtained by converting customer service call records; analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model; and analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

It should be noted that, each module in the interface interaction device may be a program module (for example, a set of program instructions for implementing a specific function), or may be a hardware module, and for the latter, it may be represented by the following form, but is not limited thereto: the expression forms of the modules are all a processor, or the functions of the modules are realized by one processor.

The foregoing embodiment numbers of the present application are merely for describing, and do not represent advantages or disadvantages of the embodiments.

In the foregoing embodiments of the present application, the descriptions of the embodiments are emphasized, and for a portion of this disclosure that is not described in detail in this embodiment, reference is made to the related descriptions of other embodiments.

In the several embodiments provided in the present application, it should be understood that the disclosed technology content may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of units may be a logic function division, and there may be another division manner in actual implementation, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be essentially or a part contributing to the related art or all or part of the technical solution may be embodied in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The foregoing is merely a preferred embodiment of the present application and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present application and are intended to be comprehended within the scope of the present application.

Claims (10)

1. A method for classifying text of a customer service telephone conversation, comprising:

obtaining a text to be classified, wherein the text to be classified is obtained by converting customer service call records;

analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model;

and analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

2. The method of claim 1, wherein analyzing the text to be classified with a first classification model and a second classification model to obtain a first prediction result and a second prediction result, respectively, comprises:

and sequentially classifying the text to be classified by adopting the first classification model to obtain the first prediction result, wherein the first prediction result comprises: the classification result of the text to be classified under three levels, wherein the classification result of each level is a sub-category of the classification result of the last level;

dividing the text to be classified into a first text set and a second text set by adopting the second classification model, and determining the second prediction result according to the first text set and the second text set, wherein the second prediction result comprises: and the first text set is formed by splitting the text to be classified into a plurality of clauses, and the second text set is formed by splitting the text to be classified into a plurality of segmentations.

3. The method of claim 2, wherein determining the second prediction result from the first set of text and the second set of text comprises:

extracting features of the first text set and the second text set by adopting the second classification model to obtain a first vector set and a second vector set, fusing vectors in the first vector set to obtain a first fused vector, and fusing vectors in the second vector set to obtain a second fused vector;

and respectively analyzing the first fusion vector and the second fusion vector by adopting the second classification model to obtain the second prediction result.

4. The method of claim 1, wherein the first classification model is trained by:

obtaining a training dataset comprising: a plurality of levels of classification labels;

training a plurality of levels of classification tasks in the first classification model by using the training data set until the first classification model converges, wherein the plurality of levels of classification tasks share the same loss function.

5. The method of claim 1, wherein fusing the first predictor and the second predictor to obtain a target predictor comprises:

acquiring weights of the first prediction result and the second prediction result;

and carrying out weighted summation on the first predicted result and the second predicted result to obtain the target predicted result.

6. The method according to claim 1, wherein analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified comprises:

determining classification results of a plurality of levels indicated by the target prediction result as a plurality of levels of a preset search tree, taking the classification result of the highest level in the plurality of levels as a root node of the preset search tree, and taking the classification results of other levels in the plurality of levels as child nodes under the root node;

sequentially selecting the next level node with highest prediction probability from the root node until the leaf node of the preset search tree is a classification node of the lowest level in the multiple levels;

and determining classification results corresponding to all nodes from the root node to the leaf node as the final classification result.

7. The method of claim 6, wherein sequentially selecting the next layer node with the highest prediction probability from the root node until the leaf node of the preset search tree comprises:

determining a plurality of node heuristic cost values in each level of the preset search tree by adopting a preset heuristic function, wherein the heuristic cost values are the prediction probabilities of the plurality of nodes;

deleting the nodes with the heuristic cost value smaller than a preset threshold value in each level to obtain a plurality of candidate nodes in each level;

and selecting the candidate node with the highest heuristic cost value from the plurality of candidate nodes of each hierarchy in turn as the next-layer node until the leaf node of the preset search tree.

8. A customer service telephone conversation text classification device, comprising:

the acquisition module is used for acquiring texts to be classified, wherein the texts to be classified are obtained by converting customer service call records;

the prediction module is used for respectively analyzing the text to be classified by adopting a first classification model and a second classification model to obtain a first prediction result and a second prediction result, and fusing the first prediction result and the second prediction result to obtain a target prediction result, wherein the target prediction result is used for indicating the classification result of the text to be classified, and the characteristic extraction mode of the first classification model is different from that of the second classification model;

and the classification module is used for analyzing the classification result indicated by the target prediction result to obtain a final classification result of the text to be classified.

9. A non-volatile storage medium, wherein a program is stored in the non-volatile storage medium, and wherein the program, when executed, controls a device in which the non-volatile storage medium is located to perform the customer service telephone conversation text classification method of any one of claims 1 to 7.

10. A computer device, comprising: a memory and a processor for executing a program stored in the memory, wherein the program is executed to perform the customer service telephone conversation text classification method of any of claims 1 to 7.