Abstract generation method and device, storage medium and computer device

Abstract

The application provides a summary generation method and device, a storage medium and a computer device. The method comprises the following steps: obtaining a work order classification model, wherein the work order classification model is a classification model for outputting a work order category based on input text; obtaining a work order summary text to be generated; inputting the work order summary text into the work order classification model, and extracting an overall text vector and each sentence text vector of the work order summary text when an encoding layer of the work order classification model generates the overall text vector and each sentence text vector of the work order summary text respectively; determining a summary sentence in each sentence of the work order summary text according to the overall text vector and each sentence text vector, and the summary sentence is used as a summary text of the work order summary text. The summary text of the work order summary text is generated by using the scheme of the application, so that the text reading amount of a customer service personnel in the process of sorting and retrieving work orders can be reduced, and the customer service response efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a summary generation method and device, a storage medium and a computer device. BACKGROUND

[0002] In an e-commerce customer service system, in order to facilitate the tracking, analysis and response of customer demands by artificial customer service, and to avoid repeated communication with customers, a work order corresponding to a customer service dialogue is generated each time a customer service dialogue with a customer is completed. The work order can include a work order summary and a work order type. The work order summary is used to record the demands raised by the customer in the dialogue and the replies to the demands by the customer service, and the work order type refers to the most core demand type in the dialogue. In this way, the artificial customer service following up can understand the previous communication by checking the work order summary generated previously.

[0003] However, considering that one dialogue can generate multiple work order summaries and one customer can generate multiple work orders, the artificial customer service following up can need to check all the texts of multiple work order summaries in order to understand the demands raised by the customer in the previous communication, which reduces the response efficiency of the artificial customer service. In order to improve the response efficiency of the customer service, it is necessary to provide a scheme capable of generating a text summary of the work order summary, so as to reduce the text reading amount of the artificial customer service. SUMMARY

[0004] The present application aims to at least solve one of the above technical defects, in particular, the technical defect of low response efficiency of the customer service caused by large text reading amount in the prior art.

[0005] In a first aspect, the present application provides a summary generation method, which comprises:

[0006] obtaining a work order classification model, the work order classification model being a classification model for outputting a work order category based on input text;

[0007] obtaining a work order summary text to be generated as a summary;

[0008] inputting the work order summary text into the work order classification model, and extracting an overall text vector and each sentence text vector of the work order summary text when an encoding layer of the work order classification model generates the overall text vector and each sentence text vector of the work order summary text respectively;

[0009] determining a summary sentence in each sentence of the work order summary text according to the overall text vector and each sentence text vector, the summary sentence being used as a summary text of the work order summary text.

[0010] In one of the embodiments, the step of determining the summary sentence in each of the sub-sentences of the work order summary text according to the overall text vector and each of the sub-sentence text vectors comprises:

[0011] calculating a similarity score between each of the sub-sentence text vectors and the overall text vector respectively;

[0012] determining the summary sentence in each of the sub-sentences of the work order summary text according to each of the similarity scores.

[0013] In one of the embodiments, the step of determining the summary sentence in each of the sub-sentences of the work order summary text according to each of the similarity scores comprises:

[0014] for each of the similarity scores, if the similarity score is greater than a preset score threshold, the sub-sentence corresponding to the similarity score is taken as the summary sentence.

[0015] In one of the embodiments, the step of calculating a similarity score between each of the sub-sentence text vectors and the overall text vector respectively comprises:

[0016] for each of the sub-sentence text vectors, calculating a cosine of angle value between the sub-sentence text vector and the overall text vector, and taking the cosine of angle value as the similarity score between the sub-sentence text vector and the overall text vector.

[0017] In one of the embodiments, the step of obtaining the work order classification model comprises:

[0018] obtaining an initial model and a plurality of sets of training data, each of the training data comprising customer service dialogue text and a pre-labeled work order category;

[0019] inputting the customer service dialogue text in each of the training data into the initial model to obtain each of the initial classifications output by the initial model, and iteratively training the initial model according to each of the initial classifications and the work order category in each of the training data until a preset training completion condition is met and the work order classification model is obtained.

[0020] In one of the embodiments, the method further comprises:

[0021] extracting each of the summary sentences from the work order summary text, and pushing each of the summary sentences to an artificial customer service.

[0022] In a second aspect, the present application provides an abstract generation device, which comprises:

[0023] The classification model acquisition module is configured to acquire a work order classification model, the work order classification model being a classification model that outputs a work order category based on input text.

[0024] The text acquisition module is configured to acquire a work order summary text for which an abstract is to be generated.

[0025] The vector extraction module is configured to input the work order summary text into the work order classification model, and extract an overall text vector and each sentence text vector of the work order summary text when the work order classification model generates the overall text vector and each sentence text vector of the work order summary text respectively in an encoding layer of the work order classification model.

[0026] The abstract generation module is configured to determine an abstract sentence in each sentence of the work order summary text according to the overall text vector and each sentence text vector, the abstract sentence being used as an abstract text of the work order summary text.

[0027] In one of the embodiments, the abstract generation module includes:

[0028] The similarity calculation unit is configured to calculate a similarity score between each sentence text vector and the overall text vector respectively.

[0029] The abstract sentence determination unit is configured to determine an abstract sentence in each sentence of the work order summary text according to each similarity score.

[0030] In a third aspect, an embodiment of the present application provides a storage medium, the storage medium storing computer readable instructions, the computer readable instructions being executed by one or more processors to cause the one or more processors to perform the steps of the abstract generation method according to any of the embodiments.

[0031] In a fourth aspect, an embodiment of the present application provides a computer device, the computer device including one or more processors and a memory.

[0032] The memory stores computer readable instructions, the computer readable instructions being executed by the one or more processors to perform the steps of the abstract generation method according to any of the embodiments.

[0033] In the abstract generation method, device, storage medium and computer device, the trained work order classification model can be used to encode the work order summary text to be generated to obtain an overall text vector and each sentence vector of the work order summary text. Since the work order classification model is a model that outputs a work order category based on an input text, the vector corresponding to the sentence strongly associated with the finally output work order category and the overall text vector of the input text in each vector output by the encoding layer of the work order classification model are more similar, so that the work order classification model outputs the work order category based on each vector output by the encoding layer. The application uses this characteristic of the work order classification model to extract the overall text vector and each sentence vector output by the encoding layer of the model after inputting the work order summary text into the work order classification model, and determines the abstract sentence of the work order summary text based on the overall text vector and each sentence vector, to obtain the abstract text of the work order summary text. In this way, the customer service personnel can know the key content of the work order summary text without browsing the full text of the work order summary text, thereby reducing the text reading amount of the customer service personnel when sorting and retrieving the work order, and improving the customer service response efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0035] Figure 1 A flowchart of the abstract generation method in one embodiment;

[0036] Figure 2 A structural diagram of the abstract generation device in one embodiment;

[0037] Figure 3 A structural diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0039] In one embodiment, the present application provides an abstract generation method. The following embodiments are described by taking the application of the method to a computer device as an example. It should be understood that the computer device described herein refers to a device with data processing function, which can be but is not limited to a notebook computer, a desktop computer, a single server, a server cluster composed of multiple servers, etc. As shown in Figure 1 The abstract generation method of the present application can include the following steps:

[0040] S102: Obtain a work order classification model, which is a classification model for outputting a work order category based on input text.

[0041] In actual application, a work order needs to be generated every time a customer service conversation with a customer is completed. In one customer service conversation, the customer can raise multiple demands. In this case, the work order summary in the work order needs to record each demand raised by the customer in this conversation and the reply of the customer service to each demand. In addition, a work order type needs to be determined according to the most core demand among the demands, so that one work order corresponds to one work order type. For example, the customer inquires about product details, discount details and raises a return in one customer service conversation, and the work order type corresponding to the work order of this conversation can be the return type.

[0042] In this step, the work order classification model refers to a model that can automatically determine the work order type according to the input text. In one example, the input text of the work order classification model can be the customer service conversation text or the work order summary text. The customer service conversation text refers to the conversation record generated when the customer communicates with the artificial customer service. The work order summary text can be a communication summary obtained from the conversation record, such as “the user inquired about the warehouse delivery problem multiple times, and a delivery acceleration order has been issued. The warehouse feedback is that the delivery needs to be delayed for three days due to epidemic control. The user has been informed of the situation and has accepted it”.

[0043] The work order classification model of the present application can include an encoding layer, which is used to calculate the overall text vector as a whole from the input text, and is also used to determine each sentence of the input text and calculate the corresponding sentence text vector of each sentence. In one embodiment, the overall text vector and each sentence text vector can be in Embedding format.

[0044] For example, when the input text is “the user has accepted the compensation plan and has informed the warehouse to accelerate delivery”, the overall text vector refers to the first vector obtained by encoding the “the user has accepted the compensation plan and has informed the warehouse to accelerate delivery” by the encoding layer, and each sentence text vector refers to the second vector obtained by encoding the sentence “the user has accepted the compensation plan” by the encoding layer and the third vector obtained by encoding the sentence “has informed the warehouse to accelerate delivery” by the encoding layer.

[0045] In one of the embodiments, S102 can include the following steps:

[0046] Step A1: obtaining an initial model and multiple sets of training data, each set of the training data including customer service dialogue text and a pre-labeled ticket category;

[0047] Step A2: inputting the customer service dialogue text in each set of the training data into the initial model respectively to obtain each initial classification output by the initial model, and performing iterative training on the initial model according to each initial classification and the ticket category in each set of the training data until a preset training completion condition is met and the ticket classification model is obtained.

[0048] Specifically, when constructing the training data of the existing extractive summary text generation model, it is necessary to manually label whether each dialogue of the dialogue text to be extracted as a summary sentence to obtain the labeled data. The training data is used for model training to obtain the final extractive summary text generation model. As can be seen, the prior art needs to consume a large amount of time cost and labor cost when constructing the training data.

[0049] In the present application, the computer device can obtain an untrained initial model and multiple sets of training data, each set of the training data including customer service dialogue text and a pre-labeled ticket category, which can be labeled by an artificial customer service after communication. The computer device can perform iterative training on the initial model using multiple sets of training data until the initial model meets a preset training completion condition, thereby obtaining a ticket classification model that can be used to accurately determine the ticket category.

[0050] In this way, on the one hand, the ticket category is used as the labeled data, and it is not necessary to manually label whether each dialogue of the customer service dialogue text needs to be a summary sentence, thereby greatly reducing the labor cost and time cost consumed when constructing the training data. On the other hand, the customer service dialogue text is used as the input data of the model. Since the text content of the customer service dialogue text is more abundant than that of the ticket summary text, the use of the ticket classification model trained based on the customer service dialogue text to calculate the overall text vector and each sentence text vector can make the sentence text vector corresponding to the sentence strongly associated with the final output ticket category more similar to the overall text vector of the input text, so as to facilitate more accurate determination of the ticket category in the subsequent process.

[0051] S104: obtaining a ticket summary text to be generated.

[0052] In one of the embodiments, the computer device can obtain the corresponding ticket summary text in the server according to the dialogue number corresponding to the current dialogue and / or the customer identifier corresponding to the customer.

[0053] S106: input the work order summary text into the work order classification model, and extract the overall text vector and each of the sentence text vectors when the work order classification model generates the overall text vector and each of the sentence text vectors respectively at an encoding layer of the work order classification model.

[0054] Specifically, the computer device can input the work order summary text for which an abstract is to be generated into the work order classification model, so that the encoding layer of the work order classification model can generate the overall text vector of the work order summary text as a whole. In addition, the encoding layer of the work order classification model can also divide the work order summary text into sentences and calculate the sentence text vector corresponding to each sentence to obtain each of the sentence text vectors of the work order summary text. It can be understood that the encoding layer can divide the text in any manner in this paper, which is not limited in the present application. In an example, the work order classification model can implement text cutting based on preset special symbols, including but not limited to “,” “.” “,” “;” and the like.

[0055] When the encoding layer of the work order classification model generates the overall text vector and each of the sentence text vectors of the work order summary text, the computer device can extract the overall text vector and each of the sentence text vectors from the work order classification model, so as to convert the work order summary text for which an abstract is to be generated into each of the vectors by using the encoding capability of the work order classification model.

[0056] S108: determining an abstract sentence in each of the sentences of the work order summary text as an abstract text of the work order summary text according to the overall text vector and each of the sentence text vectors.

[0057] Since the work order classification model is a model that outputs a work order category based on input text, the vector corresponding to the sentence strongly associated with the finally output work order category and the overall text vector of the input text are more similar in each of the vectors output by the encoding layer of the work order classification model, so as to facilitate the work order classification model to output the work order category based on each of the vectors output by the encoding layer. Since the work order category is the most core type of appeal, the sentence strongly associated with the finally output work order category is associated with the most core appeal raised by the customer. In other words, the more relevant the sentence to the most core appeal raised by the customer in each of the sentence text vectors of the work order summary text, the more similar the corresponding sentence text vector to the overall text vector. Therefore, the computer device can determine an abstract sentence in each of the sentences of the work order summary text as an abstract text based on the overall text vector and each of the sentence text vectors of the work order summary text.

[0058] It can be understood that the application can adopt any manner to implement S108, for example, clustering can be performed according to the overall text vector and each sub-sentence text vector, and the summary sub-sentence is determined based on the clustering result. In one embodiment, S108 can include the following steps:

[0059] Step B1: calculating the similarity score between each of the sub-sentence text vectors and the overall text vector, respectively;

[0060] Step B2: determining the summary sub-sentence in each of the sub-sentences of the work order summary text according to each of the similarity scores.

[0061] Specifically, for each sub-sentence text vector, the computer device can calculate the similarity score between the sub-sentence text vector and the overall text vector. It can be understood that the application can adopt any manner to calculate the similarity score between two vectors. In one example, the cosine similarity can be used as the similarity score between two vectors. That is, for each of the sub-sentence text vectors, the cosine value of the included angle between the sub-sentence text vector and the overall text vector can be calculated, and the calculated cosine value of the included angle is taken as the similarity score between the sub-sentence text vector and the overall text vector. In this way, the calculation manner of the similarity score can be simplified, thereby improving the summary generation efficiency.

[0062] After obtaining the similarity score corresponding to each sub-sentence text vector, the computer device can determine the summary sub-sentence in each sub-sentence according to each similarity score. In one embodiment, step B2 can include: for each of the similarity scores, if the similarity score is greater than a preset score threshold, the sub-sentence corresponding to the similarity score is taken as the summary sub-sentence. In other words, the computer device can take the sub-sentence corresponding to the similarity score greater than the score threshold as the summary sub-sentence. It can be understood that the score threshold herein can be determined according to actual conditions, and the application does not make specific limitation thereto.

[0063] The application determines the summary sub-sentence based on the vector similarity between each sub-sentence text and the overall text, which has the advantages of simple implementation and flexible prediction. Compared with the traditional rule-based method, the scheme of the application can improve the recall rate of the key sentence.

[0064] In one embodiment, the summary generation method of the application can further include the steps of: extracting each of the summary sub-sentences from the work order summary text, and pushing each of the summary sub-sentences to the artificial customer service. In this way, the artificial customer service can quickly understand the core demands raised by the customer in the previous communication and respond accordingly, thereby reducing the response efficiency of the artificial customer service.

[0065] The application extracts the text summary of the dialogue data based on a smart method, without manually labeling each sentence of the dialogue, and only needs to use the work order category information of the connected agent to fill in the simple dialogue text summary information.

[0066] The application can use the trained work order classification model to encode the work order summary text to be generated to obtain the overall text vector and the respective sentence text vector of the work order summary text. Since the work order classification model is a model that outputs the work order category based on the input text, the vector corresponding to the sentence that is strongly associated with the finally output work order category and the overall text vector of the input text are more similar in the respective vectors output by the encoding layer of the work order classification model, so that the work order classification model outputs the work order category based on the respective vectors output by the encoding layer. The application uses this characteristic of the work order classification model to extract the overall text vector and the respective sentence vector output by the encoding layer of the model after inputting the work order summary text into the work order classification model, and determines the summary sentence of the work order summary text based on the overall text vector and the respective sentence vector, to obtain the summary text of the work order summary text. In this way, the customer service personnel can know the key content of the work order summary text without browsing the full text of the work order summary text, thereby reducing the text reading amount and improving the customer service response efficiency.

[0067] The summary generation device provided by the embodiment of the application is described below, and the summary generation device described below can be correspondingly referred to the summary generation method described above.

[0068] In one embodiment, the application provides a summary generation device 200. As shown in the figure, the device 200 includes a classification model acquisition module 210, a text acquisition module 220, a vector extraction module 230, and a summary generation module 240. Wherein: Figure 2

[0069] The classification model acquisition module 210 is configured to acquire a work order classification model, and the work order classification model is a classification model that outputs a work order category based on an input text;

[0070] The text acquisition module 220 is configured to acquire a work order summary text to be generated;

[0071] The vector extraction module 230 is configured to input the work order summary text into the work order classification model, and extract the overall text vector and the respective sentence text vector when the work order classification model generates the overall text vector and the respective sentence text vector of the work order summary text in the encoding layer;

[0072] The summary generation module 240 is configured to determine a summary sentence in each sentence of the work order summary text according to the overall text vector and the respective sentence text vector, and the summary sentence is used as the summary text of the work order summary text.​

[0073] In an embodiment, the summary generation module 240 comprises a similarity calculation unit and a summary sentence determination unit. Wherein:

[0074] The similarity calculation unit is configured to calculate a similarity score between each of the sentence text vectors and the overall text vector respectively;

[0075] The summary sentence determination unit is configured to determine a summary sentence from each of the sentences of the work summary text according to the similarity scores.

[0076] In an embodiment, the summary sentence determination unit is configured to, for each of the similarity scores, if the similarity score is greater than a preset score threshold, take the sentence corresponding to the similarity score as the summary sentence.

[0077] In an embodiment, the similarity calculation unit is configured to, for each of the sentence text vectors, calculate a cosine of angle value between the sentence text vector and the overall text vector, and take the cosine of angle value as the similarity score between the sentence text vector and the overall text vector.

[0078] In an embodiment, the classification model acquisition module 210 comprises a data acquisition unit and a training unit.

[0079] Wherein:

[0080] The data acquisition unit is configured to acquire an initial model and a plurality of sets of training data, each set of the training data comprising customer service dialogue text and a pre-labeled work order category;

[0081] The training unit is configured to input the customer service dialogue text in each set of the training data into the initial model respectively to obtain each initial classification output by the initial model, and perform iterative training on the initial model according to each of the initial classifications and the work order category in each set of the training data until a preset training completion condition is met and the work order classification model is obtained.

[0082] In an embodiment, the summary generation apparatus 200 of the present application further comprises a pushing module. The pushing module is configured to extract each of the summary sentences from the work summary text and push each of the summary sentences to an artificial customer service.

[0083] In an embodiment, the present application further provides a storage medium having computer readable instructions stored therein, the computer readable instructions being executed by one or more processors to cause the one or more processors to perform the steps of the summary generation method according to any one of the above embodiments.

[0084] In one embodiment, the present application also provides a computer device having computer readable instructions stored therein, which, when executed by one or more processors, cause the one or more processors to perform the steps of the summary generation method of any of the above embodiments.

[0085] Schematically, Figure 3 An internal structure schematic diagram of a computer device provided by an embodiment of the present application is shown in FIG. 9. Referring to FIG. 9, Figure 3 The computer device 900 includes a processing component 902, which further includes one or more processors, and a memory resource represented by a memory 901 for storing instructions executable by the processing component 902, such as an application program. The application program stored in the memory 901 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 902 is configured to execute the instructions to perform the steps of the summary generation method of any of the above embodiments.

[0086] The computer device 900 can also include a power supply component 903 configured to perform power management of the computer device 900, a wired or wireless network interface 904 configured to connect the computer device 900 to a network, and an input / output (I / O) interface 905. The computer device 900 can operate based on an operating system stored in the memory 901, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0087] Those skilled in the art can understand, Figure 3 The structure shown in FIG. 9 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0088] Finally, it should be noted that the terms "first" and "second", and the like, herein do not denote any order, quantity, combination or importance, but are used to identify one element from another, and do not imply that the specific identities thereof are essential or that the identities are chronological or related in their occurrence. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element. Also, "a", "an", "the", and "said" are used to refer to one or more than one (i.e., to "at least one") of the referenced elements, unless otherwise specified. A plurality also means two or more, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of one or more of the associated listed items.

[0089] The various embodiments described in this specification are presented by way of example, and are not intended to limit the scope of the application. Each embodiment is presented in a way that emphasizes the differences between the embodiments and the other embodiments, but the embodiments can be combined according to the needs of the user, and the same or similar parts are cross-referenced.

[0090] The above description of disclosed embodiments provides enabling disclosure sufficient for one of ordinary skill in the art to practice or use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An abstract generation method characterized by comprising: The method comprises: obtaining a work order classification model, the work order classification model being a classification model for outputting a work order category based on input text; obtaining a work order summary text to be generated into an abstract; inputting the work order summary text into the work order classification model, and extracting an overall text vector and each sub-sentence text vector of the work order summary text when the work order classification model generates the overall text vector and each sub-sentence text vector respectively at an encoding layer of the work order classification model; determining an abstract sub-sentence in each sub-sentence of the work order summary text according to the overall text vector and each sub-sentence text vector, the abstract sub-sentence being used as an abstract text of the work order summary text; wherein the step of obtaining the work order classification model comprises: obtaining an initial model and a plurality of sets of training data, each set of the training data comprising customer service dialogue text and a pre-labeled work order category; inputting the customer service dialogue text in each set of the training data into the initial model respectively to obtain each initial classification output by the initial model, and iteratively training the initial model according to each initial classification and the work order category in each set of the training data until a preset training completion condition is met and the work order classification model is obtained.

2. The abstract generation method according to claim 1, characterized by, The step of determining an abstract sub-sentence in each sub-sentence of the work order summary text according to the overall text vector and each sub-sentence text vector comprises: calculating a similarity score between each sub-sentence text vector and the overall text vector respectively; determining an abstract sub-sentence in each sub-sentence of the work order summary text according to each similarity score.

3. The abstract generation method according to claim 2, characterized by, The step of determining an abstract sub-sentence in each sub-sentence of the work order summary text according to each similarity score comprises: for each similarity score, if the similarity score is greater than a preset score threshold, the sub-sentence corresponding to the similarity score is taken as the abstract sub-sentence.

4. The abstract generation method according to claim 2 or 3, characterized by, The step of calculating a similarity score between each sub-sentence text vector and the overall text vector comprises: for each sub-sentence text vector, calculating a cosine of angle value between the sub-sentence text vector and the overall text vector, and taking the cosine of angle value as the similarity score between the sub-sentence text vector and the overall text vector.

5. The abstract generation method according to any one of claims 1 to 3, characterized by, The method further comprises: extracting each abstract sub-sentence from the work order summary text, and pushing each abstract sub-sentence to an artificial customer service.

6. An abstract generation apparatus characterized by comprising: The device comprises: a classification model obtaining module configured to obtain a work order classification model, the work order classification model being a classification model for outputting a work order category based on input text; specifically comprising: obtaining an initial model and a plurality of sets of training data, each set of the training data comprising customer service dialogue text and a pre-labeled work order category; inputting the customer service dialogue text in each set of the training data into the initial model respectively to obtain each initial classification output by the initial model, and iteratively training the initial model according to each initial classification and the work order category in each set of the training data until a preset training completion condition is met and the work order classification model is obtained; The text acquisition module is configured to acquire a work order summary text to be generated as an abstract; The vector extraction module is configured to input the work order summary text into the work order classification model, and extract an overall text vector and each sentence text vector of the work order summary text when the work order classification model generates the overall text vector and each sentence text vector respectively in an encoding layer of the work order classification model. The abstract generation module is configured to determine an abstract sentence in each sentence of the work order summary text according to the overall text vector and each sentence text vector, and the abstract sentence is used as an abstract text of the work order summary text.

7. The abstract generation apparatus according to claim 6, wherein The abstract generation module includes: A similarity calculation unit is configured to calculate a similarity score between each sentence text vector and the overall text vector respectively; An abstract sentence determination unit is configured to determine an abstract sentence in each sentence of the work order summary text according to each similarity score.

8. A storage medium characterized by: The storage medium stores computer readable instructions, and the computer readable instructions are executed by one or more processors to cause the one or more processors to perform the steps of the abstract generation method according to any one of claims 1 to 5.

9. A computer device, comprising: It includes: One or more processors and a memory; The memory stores computer readable instructions, and the computer readable instructions are executed by the one or more processors to perform the steps of the abstract generation method according to any one of claims 1 to 5.

Images