CN117974271A

CN117974271A - Event summary generation and semantic recognition event summary model training method and device

Info

Publication number: CN117974271A
Application number: CN202410271184.9A
Authority: CN
Inventors: 韩振磊
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2024-03-11
Filing date: 2024-03-11
Publication date: 2024-05-03

Abstract

The disclosure provides an event summary generation method and device and a semantic recognition event summary model training method and device. One embodiment of the event summary generation method includes: when customer service communicates with a customer, acquiring commodity order information related to communication; detecting whether a recommended result of the event summary exists or not based on commodity order information; acquiring recording information of a call in response to detecting a recommendation result without an event summary; and identifying an event summary model based on commodity order information, recording information and pre-trained semantics to obtain a recommendation result of the event summary. This embodiment improves the accuracy of the event summary generation.

Description

Event summary generation and semantic recognition event summary model training method and device

Technical Field

The present disclosure relates to the field of computer technology, and in particular, to the field of artificial intelligence technology, and more particularly, to an event summary generating method and apparatus, a semantic recognition event summary model training method and apparatus, an electronic device, and a computer readable storage medium.

Background

The intelligent event summary recommendation of customer service in the E-commerce field is as follows: in the process of line listening of customer service or after the process is finished, a recommendation engine recommends an event summary for line listening of the current line listening to select customer service according to the content, event summary log, customer information, commodity information, order information and the like of the customer service communicated with a customer phone.

In an actual scene, because event summaries are various, customer service needs to take a long time to select the event summaries of the events, in general, an executing body extracts some characteristic information with a relatively tight relation with the event probability from commodity order information, and predicts and recommends the event summaries of the listening line in the process of customer service wiring by using a big data statistics, statistics learning or deep learning method, but the accuracy of recommendation is low due to lack of characteristic interaction, and the accuracy of event summary recommendation is possibly low in response efficiency under multiple concurrent scenes.

Disclosure of Invention

The embodiment of the disclosure provides an event summary generation method and device, a semantic recognition event summary model training method and device, electronic equipment and a computer readable storage medium.

In a first aspect, embodiments of the present disclosure provide an event summary generating method, the method including: when customer service communicates with a customer, acquiring commodity order information related to communication; detecting whether a recommended result of the event summary exists or not based on commodity order information; acquiring recording information of a call in response to detecting a recommendation result without an event summary; based on commodity order information, recording information and a pre-trained semantic recognition event summary model, a recommendation result of the event summary is obtained, and the semantic recognition event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories.

In some embodiments, the commodity order information includes: customer information, merchant information, merchandise information, and order information; the detecting whether the recommended result has the event summary based on the commodity order information includes: determining characteristics of customer information, merchant information, commodity information and order information; and detecting whether the recommended result of the event summary exists or not based on the consistent characteristics in the client information, the merchant information, the commodity information and the order information and a preset rule system.

In some embodiments, identifying the event summary model based on the commodity order information, the recording information and the pre-trained semantics, and obtaining the recommended result of the event summary includes: based on the recording information, obtaining text semantic features; extracting commodity order features based on commodity order information; inputting text semantic features and commodity order features into a pre-trained semantic recognition event summary model to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary model; and obtaining a recommendation result of the event summary based on the probabilities of the event summaries.

In some embodiments, the obtaining the recommended result of the event summary based on the probabilities of the event summaries includes: and ordering the probabilities of all event summaries in the event summaries of the preset types in a descending order to obtain the probability of the pre-set bit, and taking the event summary corresponding to the probability of the pre-set bit as a recommendation result.

In some embodiments, the obtaining the recommended result of the event summary based on the probabilities of the event summaries includes: determining thresholds of various event summaries based on historical data of the various event summaries; based on the probability of each event summary, detecting whether each event summary meets the requirement of each threshold; and responding to the requirement that the event summary meets the threshold value, and taking the event summary as a recommendation result.

In a second aspect, embodiments of the present disclosure provide a semantic recognition event summary model training method, the method comprising: obtaining a sample set, the sample set comprising: at least one sample, each sample comprising: sample text semantic features, and features of merchandise order information corresponding to the sample text semantic features; acquiring a semantic identification event summary network, wherein the semantic identification event summary network is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories; the following training steps are performed: selecting a sample from the sample set; inputting the sample into a semantic recognition event summary network to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary network; and responding to the semantic recognition event summary network to meet the training completion condition, and obtaining a semantic recognition event summary model corresponding to the corresponding semantic recognition event summary network.

In some embodiments, the semantic recognition event summary network described above includes: the system comprises a text semantic feature network and a discrete feature network, wherein the text semantic feature network is used for representing the corresponding relation between text semantic features and various event summaries in preset categories; the discrete feature network is used for representing the corresponding relation between the features of commodity order information and various event summaries in the preset category.

In a third aspect, embodiments of the present disclosure provide an event summary generating apparatus, including: an information acquisition unit configured to acquire commodity order information related to a call when a customer service calls with a customer; a detection unit configured to detect whether a recommended result of the event summary is present or not based on the commodity order information; a recording acquisition unit configured to acquire recording information of a call in response to detecting a recommendation result without an event summary; the result obtaining unit is configured to obtain a recommendation result of the event summary based on commodity order information, recording information and a pre-trained semantic identification event summary model, wherein the semantic identification event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories.

In some embodiments, the commodity order information includes: customer information, merchant information, merchandise information, and order information; the above detection unit is further configured to: determining characteristics of customer information, merchant information, commodity information and order information; and detecting whether the recommended result of the event summary exists or not based on the consistent characteristics in the client information, the merchant information, the commodity information and the order information and a preset rule system.

In some embodiments, the above-described result obtaining unit is further configured to: based on the recording information, obtaining text semantic features; extracting commodity order features based on commodity order information; inputting text semantic features and commodity order features into a pre-trained semantic recognition event summary model to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary model; and obtaining a recommendation result of the event summary based on the probabilities of the event summaries.

In some embodiments, the above-described result obtaining unit is further configured to: and ordering the probabilities of all event summaries in the event summaries of the preset types in a descending order to obtain the probability of the pre-set bit, and taking the event summary corresponding to the probability of the pre-set bit as a recommendation result.

In some embodiments, the above-described result obtaining unit is further configured to: determining thresholds of various event summaries based on historical data of the various event summaries; based on the probability of each event summary, detecting whether each event summary meets the requirement of each threshold; and responding to the requirement that the event summary meets the threshold value, and taking the event summary as a recommendation result.

In a fourth aspect, embodiments of the present disclosure provide a semantic recognition event summary model training apparatus, the apparatus comprising: a sample acquisition unit configured to acquire a sample set including: at least one sample, each sample comprising: sample text semantic features, and features of merchandise order information corresponding to the sample text semantic features; the network acquisition unit is configured to acquire a semantic recognition event summary network, wherein the semantic recognition event summary network is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories; a selecting unit configured to select a sample from a sample set; the input unit is configured to input the sample into the semantic recognition event summary network to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary network; the model obtaining unit is configured to obtain a semantic recognition event summary model corresponding to the semantic recognition event summary network in response to the semantic recognition event summary network meeting the training completion condition.

In a fifth aspect, embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device having one or more programs stored thereon; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as described in any embodiment of the first or second aspects.

In a sixth aspect, embodiments of the present disclosure provide a computer readable medium having stored thereon a computer program which when executed by a processor implements a method as described in any of the embodiments of the first or second aspects.

The embodiment of the disclosure provides a method and a device for generating event summary, firstly, when customer service and a customer call, commodity order information related to the call is obtained; secondly, based on commodity order information, detecting whether a recommendation result of the event summary exists or not; thirdly, acquiring recording information of the call in response to detecting a recommendation result without the event summary; finally, based on commodity order information, recording information and a pre-trained semantic identification event summary model, a recommendation result of the event summary is obtained, and the semantic identification event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories. Therefore, when the recommendation result of the event summary cannot be detected through commodity order information, the recommendation result of the event summary is obtained based on the call recording information, the summary log, the commodity order information and the pre-trained semantic recognition event summary model, and the semantic recognition event summary model is introduced to analyze the event summary on the basis that the recommendation result cannot be obtained through the commodity order information, so that a large amount of data distribution effects are achieved; by introducing the semantic recognition event summary model, the barrier for predicting the event summary only from customer service and customer call content is broken, and the accuracy of event summary prediction is improved.

Drawings

Other features, objects and advantages of the present disclosure will become more apparent upon reading of the detailed description of non-limiting embodiments, made with reference to the following drawings:

FIG. 1 is an exemplary system architecture diagram in which an embodiment of the present disclosure may be applied;

FIG. 2 is a flow chart of one embodiment of an event summary generation method according to the present disclosure;

FIG. 3 is a schematic diagram of a data structure of steps of an event summary generation method according to the present disclosure;

FIG. 4 is a flow chart of one embodiment of a semantic recognition event summary model training method according to the present disclosure;

FIG. 5 is a schematic diagram of an embodiment of an event summary generation apparatus according to the present disclosure;

FIG. 6 is a structural schematic diagram of one embodiment of a semantic recognition event summary model training apparatus according to the present disclosure;

fig. 7 is a schematic structural diagram of an electronic device suitable for use in implementing embodiments of the present disclosure.

Detailed Description

The present disclosure is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings.

It should be noted that, without conflict, the embodiments of the present disclosure and features of the embodiments may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

FIG. 1 illustrates an exemplary system architecture 100 to which the event summary generation method or the semantic recognition event summary model training method of the present disclosure may be applied.

As shown in fig. 1, the system architecture 100 may include terminals 101, 102, a network 103, a database server 104, and a server 105. The network 103 serves as a medium for providing a communication link between the terminals 101, 102, the database server 104 and the server 105. The network 103 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user 110 may interact with the server 105 via the network 103 using the terminals 101, 102 to receive or send messages or the like. The terminals 101, 102 may have various client applications installed thereon, such as model training class applications, image recognition applications, shopping class applications, payment class applications, web browsers, instant messaging tools, and the like.

The terminals 101 and 102 may be hardware or software. When the terminals 101, 102 are hardware, they may be various electronic devices with display screens, including but not limited to smartphones, tablet computers, electronic book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic video experts compression standard audio plane 3), laptop and desktop computers, and the like. When the terminals 101, 102 are software, they can be installed in the above-listed electronic devices. Which may be implemented as multiple software or software modules (e.g., to provide distributed services), or as a single software or software module. The present invention is not particularly limited herein.

Database server 104 may be a database server that provides various services. For example, a database server may have stored therein a sample set. The sample set contains a large number of samples, each of which is different, and the samples may include user features, manipulated object features, and contextual features between the user and the manipulated object. The user 110 may also select samples from the set of samples stored by the database server 104 via the terminals 101, 102.

The server 105 may also be a server providing various services, such as a background server providing support for various applications displayed on the terminals 101, 102. The background server may train the semantic recognition event summary model training model by using the samples in the sample set sent by the terminals 101 and 102, and may send the semantic recognition event summary model training model obtained by training to the terminals 101 and 102. In this way, the user may apply the generated semantic recognition event summary model training model to predict the event summary.

The database server 104 and the server 105 may be hardware or software. When they are hardware, they may be implemented as a distributed server cluster composed of a plurality of servers, or as a single server. When they are software, they may be implemented as a plurality of software or software modules (e.g., to provide distributed services), or as a single software or software module. The present invention is not particularly limited herein.

It should be noted that, the event summary generating method or the semantic recognition event summary model training method provided by the embodiments of the present disclosure is generally executed by the server 105. Accordingly, an event summary generation means or a semantic recognition event summary model training means is also typically provided in the server 105.

It should be noted that the database server 104 may not be provided in the system architecture 100 in cases where the server 105 may implement the relevant functions of the database server 104.

It should be understood that the number of terminals, networks, database servers, and servers in fig. 1 are merely illustrative. There may be any number of terminals, networks, database servers, and servers, as desired for implementation.

In the conventional technology, the current event summary recommendation mainly comprises the following three methods 1) to 3), and various methods have corresponding disadvantages:

1) Rule-based method

And when the online customer service listens to the online customer service, the recommendation engine sorts the event summary selectivity corresponding to the commodity in a large amount of historical data in a descending order, and recommends the first three event summaries to the online customer service. The method has the following defects: the new commodity has no history data, so that no method is recommended; the historical data of a large number of long-tail commodities is small, so that the recommendation accuracy is poor; the types of event summaries corresponding to commodities are many, so that the data of the preset position of the historical data selection rate is very small in proportion, and the upper limit of the accuracy is low; the customer service and customer call content feature cannot be utilized.

2) Statistical learning-based method

Features related to the event summary are mainly distribution information, invoice information, client information, commodity information, merchant information, order information and the like, are extracted from a large amount of historical data, an event summary classifier is constructed, and the classifier mainly adopts a statistical learning method. The method has the following defects: the distribution of the extracted features in the feature space is not compact enough; the classifier constructed by the statistical learning method needs to manually extract interaction characteristics; the customer service and customer call content feature cannot be utilized.

3) Deep learning-based method

And (3) identifying customer service and customer dialogue contents by using an ASR (Automatic Speech Recognition (automatic speech recognition technology) technology, acquiring customer intention by using a semantic identification technology according to the dialogue contents, and constructing an event summary recommendation engine by combining customer information, commodity information, merchant information, order information and the like. The method has the following defects: the rich customer service and customer call content information cannot be fully utilized. The category of customer intent predicted by the call content is far less than the category of event summary; the lack of interaction between the intent expressed by the customer and the merchandise, merchant, customer, order information; the deep learning method has long reasoning time.

Aiming at the problems of low event summary recommendation accuracy and long prediction time in the prior art, the present disclosure provides an event summary generation method, which adopts a pre-trained semantic recognition event summary model to obtain a recommendation result of an event summary on the basis that a recommendation result of an event summary cannot be obtained through commodity order information, so as to improve the accuracy of event summary obtaining, as shown in fig. 2, please refer to fig. 2, which shows a flow 200 of one embodiment of the event summary generation method provided by the present disclosure, the event summary generation method may include the following steps:

in step 201, when customer service communicates with customer, commodity order information related to communication is acquired.

In this embodiment, the customer service may be an online customer service or a manual customer service, and when the customer service communicates with the customer, and especially after the online customer service is connected, the execution body on which the event summary generating method is operated may communicate with the terminal (such as terminals 101 and 102 in fig. 1) to obtain the commodity order information sent by the terminal.

In this embodiment, the commodity order information is information of commodities, customers, merchants, orders related to call content, and for example, the customer information includes: customer attribute data and customer behavior data, wherein the customer attribute data includes customer age, sex, and the like. The customer behavior data includes: the sequence of items clicked by the user over a historical period. The commodity information includes: commodity name, commodity type, commodity color, commodity place of origin, relationship between commodity and customer, etc.

Step 202, based on the commodity order information, it is detected whether there is a recommendation result of the event summary.

In this embodiment, whether the recommended result of the event summary is provided may be detected in various manners, specifically, step 202 includes: acquiring historical data corresponding to commodity order information; detecting whether the historical data has an event summary corresponding to a commodity of commodity order information; if so, ordering the event summaries corresponding to the commodities of the commodity order information to obtain the recommendation results of the event summaries.

In this embodiment, after the event summary is summarized on the call contents of the customer service and the customer, the obtained content summary is the event summary closest to the current customer service and the customer call, which is selected from the event summaries of preset types, and the recommended result of the event summary may be sent to the personnel customer service, and the personnel customer service may select the event summary closest to the current call from the recommended results. Alternatively, when the recommendation does not match the current call at all, the manual customer service needs to manually select the correct event profile.

Optionally, the event summary generating method further includes: in response to detecting a recommendation with an event summary, steps 203-204 are no longer performed.

In step 203, recording information of the call is acquired in response to detecting the recommendation result without the event summary.

In this embodiment, when a customer service and a customer call, the call between the customer service and the customer call can be recorded by the recording device, so as to obtain recording information of the call.

And 204, identifying an event summary model based on commodity order information, recording information and pre-trained semantics to obtain a recommendation result of the event summary.

In this embodiment, the semantic recognition event summary model is used to characterize the corresponding relationship between the text semantic features, the features of the commodity order information and various event summaries in the preset category.

The step 204 includes: based on the recording information, obtaining an identification text; based on the identification text, obtaining text semantic features; extracting characteristics of commodity order information based on the commodity order information; inputting the identification features and the features of the commodity order information into a semantic identification event summary model to obtain an event summary output by the semantic identification event summary model; the event summary output by the semantic recognition event summary model may be probabilities of various event summaries in a preset category, descending order of probabilities of all event summaries is ordered based on probabilities of all event summaries output by the semantic recognition time summary model, and event summaries of preset bits (the preset bits may be determined based on generation requirements, for example, set to the first 3 bits) are selected as recommendation results of the event summaries.

It should be noted that, the event summary generating method of the present embodiment may be used to test a pre-trained semantic recognition event summary model. And then the semantic recognition event summary model can be continuously optimized according to the test result.

The task of the semantic recognition event summary model aims at obtaining a recommendation result of the event summary based on commodity order information and recording information, and the event summary matched with the current conversation pair can be determined through the recommendation result of the event summary, so that a preferred implementation mode is provided for generating the event summary for an execution subject, and customer service line listening efficiency is improved in an auxiliary mode.

Optionally, the event summary generating method further comprises the steps of obtaining a summary log based on the recording information; the step 204 further includes: based on the recording information, obtaining an identification text; extracting features of the identification text and the summary log to obtain text semantic features of the identification text; extracting characteristics of commodity order information based on the commodity order information; inputting the text semantic features and the features of commodity order information into a semantic identification event summary model to obtain an event summary output by the semantic identification event summary model.

As shown in fig. 3, the above scenario is: in the process of customer service line listening, a summary log is compiled according to dialogue content with customers and commodity order information, and after the summary log is compiled, an event summary of a line listening event is selected. However, in a real scene, because of various event summaries, online customer service needs to take a long time to correctly select the event summary of the event. The main purpose of the event summary generation method provided by the disclosure is as follows: the efficiency of online customer service line listening is improved, 3 event summaries are recommended to the manual customer service by the line listening recommendation engine each time, and if one of the 3 event summaries matches with the current line listening content, the manual customer service is directly selected; if all 3 event summaries do not match the current listening line content, then manual customer service needs to manually select the correct event summary.

According to the event summary generation method provided by the embodiment of the disclosure, firstly, when customer service and a customer call, commodity order information related to the call is acquired; secondly, based on commodity order information, detecting whether a recommendation result of the event summary exists or not; thirdly, acquiring recording information of the call in response to detecting a recommendation result without the event summary; finally, based on commodity order information, recording information and a pre-trained semantic identification event summary model, a recommendation result of the event summary is obtained, and the semantic identification event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories. Therefore, when the recommendation result of the event summary cannot be detected through commodity order information, the recommendation result of the event summary is obtained based on the call recording information, the summary log, the commodity order information and the pre-trained semantic recognition event summary model, and the semantic recognition event summary model is introduced to analyze the event summary on the basis that the recommendation result cannot be obtained through the commodity order information, so that a large amount of data distribution effects are achieved; by introducing the semantic recognition event summary model, the barrier for predicting the event summary only from customer service and customer call content is broken, and the accuracy of event summary prediction is improved.

In some optional implementations of the disclosure, the commodity order information includes: customer information, merchant information, merchandise information, and order information; based on the merchandise order information, detecting whether the recommended result has the event summary includes: determining characteristics of customer information, merchant information, commodity information and order information; and detecting whether the recommended result of the event summary exists or not based on the consistent characteristics in the client information, the merchant information, the commodity information and the order information and a preset rule system.

In this optional implementation manner, the preset rule system associates multiple kinds of information with consistent features (for example, the feature of the age of the customer in the customer information is consistent with the feature of the adaptive age bracket in the commodity information) with event summaries of corresponding kinds according to a preset rule, when obtaining information with features, the rule system can query the features according to the rule to obtain event summaries corresponding to the features, and the obtained event summaries corresponding to the features are recommendation results of the event summaries.

When the corresponding event summary cannot be queried from a preset rule system according to some rules, the recommendation result of the event summary is not available.

In this alternative implementation, the rule system is constructed as follows: and taking the completely consistent features in the features of the customer information, the merchant information, the commodity information and the order information as a group, wherein the completely consistent features are added into a rule system because the event summary selected by each feature is not necessarily the same, and the ratio of the sample size according to the preset position (such as the first 3 bits) in the whole group is larger than a threshold value (the threshold value description is centralized). Less than threshold value, the descriptions are more scattered, customer service opinions are inconsistent, noise data are obtained, and a rule system is not needed to be added.

In this embodiment, as shown in fig. 3, when the customer service line is turned on, the rule system recommendation is triggered immediately. If the rule system cannot complete recommendation, i.e. cannot obtain the recommendation result of the event summary, waiting for customer service to write the summary log, triggering a semantic recognition event summary model, and carrying out the recommendation result of the event summary to obtain a flow. The semantic recognition event summary model takes commodity order characteristics (discrete characteristics in fig. 3) generated by client information, merchant information, commodity information and order information, and text semantic characteristics generated by a recognition text and a summary log after voice recognition of customer service and client call content (recording information) as inputs, and outputs a recommended event summary, so that the response speed and the accuracy of event summary recommendation are greatly improved on the premise of ensuring the accuracy.

The method for generating the event summary provided by the alternative implementation mode provides a recommendation scheme of the event summary combined with a rule system, wherein the event summary of each group is counted according to the same statistics of all the customer, merchant, commodity and order information fields in historical data, if the sample ratio of the preset bits is more than or equal to 95% of the total sample amount of the group (an adjustable threshold value), the rule system is added, and the recommendation engine preferentially executes the rule system. And a large amount of data is shunted to a rule system on the premise of high accuracy, so that the reasoning time is improved.

The method for detecting the recommendation result of the event summary provided by the alternative implementation mode detects the event summary from the rule system based on the consistent features in the commodity order information, provides an alternative implementation mode for whether the recommendation result of the event summary exists or not, and improves the reliability of the recommendation result of the event summary.

Optionally, the detecting whether the recommended result has the event summary based on the commodity order information includes: determining the characteristics related to the event summaries in commodity order information, inputting the characteristics related to the event summaries into a pre-trained classification model to obtain the probability of each type of event summaries in the preset types output by the classification model, and taking the event summaries which are previously set and positioned in the probability of each type of event summaries in the preset types output by the classification model as the recommendation results of the event summaries.

In some optional implementations of the disclosure, identifying the event summary model based on the commodity order information, the recording information and the pre-trained semantics, obtaining the recommended result of the event summary includes: based on the recording information, obtaining text semantic features; extracting commodity order features based on commodity order information; inputting text semantic features and commodity order features into a pre-trained semantic recognition event summary model to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary model; and obtaining a recommendation result of the event summary based on the probabilities of the event summaries.

In this optional implementation manner, the obtaining text semantic features based on the recording information includes: performing text conversion on the recording information by adopting an audio text converter to obtain an identification text; extracting semantic features in the identification text to obtain text semantic features.

In the optional implementation manner, the semantic recognition event summary model is a model for comprehensively analyzing text semantics and commodity order features to obtain probabilities of various event summaries in preset types, and the semantic recognition event summary model is a model obtained by carrying out joint learning on the semantic recognition model and the event summary recommendation model. The semantic identification event summary model breaks through barriers that customer intention is predicted from customer service and customer communication content, and then the event summary is predicted by using customer intention, customer, merchant, commodity and order information, so that the accuracy of the event summary is improved.

In some optional implementations of the disclosure, the obtaining the recommendation result of the event summary based on the probabilities of the event summaries includes: and ordering the probabilities of all event summaries in the event summaries of the preset types in a descending order to obtain the probability of the pre-set bit, and taking the event summary corresponding to the probability of the pre-set bit as a recommendation result.

In this alternative implementation, the pre-set bit may be determined according to the actual data acquisition requirement, for example, the pre-set bit is the first 3 bits.

According to the method for obtaining the recommended result of the event summary, which is provided by the alternative implementation mode, the probability of the event summary of the preset type is ordered in a descending order, the probability of the pre-set bit is selected, the event summary corresponding to the probability of the pre-set bit is used as the recommended result, and the accuracy of obtaining the recommended result of the event summary is improved.

Optionally, the obtaining the recommendation result of the event summary based on the probabilities of the event summaries includes: and carrying out ascending order on the probabilities of all event summaries in the event summaries of the preset types to obtain the probability of the post-setting bit, and taking the event summaries corresponding to the probability of the post-setting bit as recommendation results.

Optionally, the obtaining the recommendation result of the event summary based on the probabilities of the event summaries includes: determining event summaries with the same probability in the event summaries of the preset types, and merging the event summaries with the same probability together to obtain new event summaries of the types; and sorting the probabilities of the new types of event summaries in ascending order or descending order, and selecting the event summaries with the post-setting bit probabilities or the pre-setting bit probabilities as recommendation results.

In some optional implementations of the disclosure, the obtaining the recommendation result of the event summary based on the probabilities of the event summaries includes: determining thresholds of various event summaries based on historical data of the various event summaries; based on the probability of each event summary, detecting whether each event summary meets the requirement of each threshold; and responding to the requirement that the event summary meets the threshold value, and taking the event summary as a recommendation result.

In the alternative implementation manner, whether various event summaries meet the requirement of the threshold value can be detected through a formula shown as a formula (1), such as a formula (1)Is a threshold value for various event summaries.

Where σ represents the variance in the historical data, μ represents the expected value of the historical data, ε represents the controllable threshold, X is the probability of the semantic recognition event summary model output, and C is a data processing function, e.g., rounding the data, or rounding the data, etc.

The method for obtaining the recommended result of the event summary provided by the alternative implementation mode determines thresholds of various event summaries based on the historical data of the event summary; based on the probability of various event summaries, whether the various event summaries meet the requirements of respective thresholds is detected, and when the event summaries meet the requirements of the thresholds, the event summaries are used as recommendation results. Thus, a reliable implementation is provided for obtaining the recommended results of the event summary.

One specific implementation flow of the event summary generating method provided by the present disclosure is as follows: as shown in fig. 3, after the online customer service is wired, the execution main body on which the event summary generating method is run automatically acquires commodity order information including customer information, merchant information, commodity information and order information, and triggers the rule system to perform an event summary recommending flow, at this time, if the rule system can recommend that the event summary recommending flow directly returns a recommending result, if the event summary recommending flow cannot recommend, the rule system enters a waiting mode.

Starting real-time recording when customer service begins to answer customer calls, identifying recording information, primarily knowing customer requirements in customer service and customer calls, simultaneously starting to write a summary log by the customer service, judging that the event content is communicated when the customer service clicks to save the summary log, and triggering an event summary recommending process through a pre-trained semantic identification event summary model.

The execution main body on which the event summary generation method operates collects order information such as identification text, summary log text, client information, commodity information, merchant information, logistics/invoice/work order/dispute/pay and the like of customer service and client call content.

And outputting the probability of each event summary through online prediction of a semantic recognition event summary model, sorting probability values in descending order, and selecting the first three event summaries as event summary recommendation results to be recommended to the customer service.

The manual customer service judges whether the recommended result contains the correct event summary, if so, the manual customer service directly selects, and if not, the manual customer service searches and selects the correct event summary.

In the field of electronic commerce, the event summary generation method can improve the on-line customer service line listening efficiency. In addition, by mining the features highly related to the event summary and combining the semantic recognition event summary model, the accuracy of event summary recommendation is greatly improved; and the response speed of the recommendation engine is greatly improved by using a rule system for shunting. In the test, the accuracy of the method provided by the invention is improved by 40 percentage points compared with the method based on rules, the accuracy of the recommended result is improved by 25 percentage points compared with the method based on statistical learning, the accuracy of the recommended result reaches 92%, and the average human efficiency is saved by 50 people/day; the response speed tie is improved by 50 ms/sheet.

Referring to FIG. 4, a flow 400 of one embodiment of a semantic recognition event summary model training method according to the present disclosure is shown, the semantic recognition event summary model training method comprising the steps of:

In step 401, a sample set is acquired.

In this embodiment, the sample set is a sample set collected in advance for training a semantic recognition event summary model, where samples in the sample set may be obtained by collecting sample data from the internet and performing feature extraction (for example, by a feature extractor) on the sample data, where the sample set includes at least one sample, and each sample includes a sample text semantic feature and a feature of commodity order information corresponding to the sample text semantic feature.

In this embodiment, the features of the order information of the commodity may be features of the commodity, features of the customer, features of the merchant, and features of the order in the same website or application specific content, where the features of the commodity include: the color of the commodity, the type of the commodity, the place of production of the commodity, etc.; the characteristics of the customer include: customer name, customer preference, etc., merchant characteristics include: merchant location, merchant name, merchant product, etc.

In this embodiment, the sample text voice feature is a specific feature describing the text of the commodity, for example, information such as good public praise, commodity evaluation, etc.

In this embodiment, the execution subject of the semantic recognition event summary model training method (e.g., the server 105 shown in fig. 1) may acquire a sample set in various ways. For example, the executing entity may obtain the sample set stored therein from a database server (e.g., database server 104 shown in fig. 1) through a wired connection or a wireless connection. As another example, a user may collect a sample through a terminal (e.g., terminals 101, 102 shown in fig. 1). In this way, the executing body may receive samples collected by the terminal and store the samples locally, thereby generating a sample set.

Step 402, obtaining a pre-constructed semantic recognition event summary model training network.

In this embodiment, the semantic recognition event summary network is a network combining semantic recognition and event summary recommendation. The text semantic features and the features of commodity order information are simultaneously input into the semantic recognition event summary network, and the semantic recognition time summary network simultaneously analyzes the text semantic and the features of the commodity order information to obtain a prediction result of the event summary.

At step 403, samples are selected from the sample set.

In this embodiment, the executing body may select a sample from the sample set obtained in step 401, and execute the training steps from step 404 to step 405. The selection manner and the selection number of the samples are not limited in the present application. For example, at least one sample may be randomly selected.

Step 404, inputting the sample into a semantic recognition event summary network to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary network.

In this embodiment, the semantic recognition event summary network is used to characterize the corresponding relationship between the text semantic features, the features of the commodity order information and various event summaries in the preset category.

In this embodiment, after the sample is input into the semantic recognition event summary network, a loss value of the semantic recognition event summary model training network may be calculated.

In this embodiment, in the training process of the semantic recognition event summary model training network, multiple iterative training is generally required to be performed to enable the overall loss of the semantic recognition event summary model training network to converge, each iterative training sequentially executes steps 403-404, after calculating the loss value, adjusts parameters of the semantic recognition event summary model training network based on the loss value (the loss value reflects the magnitude of the error), and then performs the next iterative training, and continues to execute steps 403-404 until the training completion condition is satisfied.

In this embodiment, in order to effectively adjust parameters of the semantic recognition event summary network, a loss function needs to be set for the semantic recognition event summary network, and the loss function can calculate errors between a forward calculation result and a true value of each iteration of the semantic recognition event summary network, so that the next training is guided to be performed in a correct direction through the errors.

It should be noted that, in the training step of the current iteration number, if the semantic recognition event summary network already meets the training completion condition, the execution subject will not input a sample into the semantic recognition event summary network, and the semantic recognition event summary network is the semantic recognition event summary model after training is completed.

And step 405, responding to the fact that the semantic recognition event summary network meets the training completion condition, and obtaining a semantic recognition event summary model corresponding to the corresponding semantic recognition event summary network.

In this embodiment, the training completion condition includes at least one of: the training iteration times of the semantic recognition event summary complex reach a preset iteration threshold; and when the change rate of the model parameters of the semantic identification event summary network is smaller than the preset threshold value, determining that the semantic identification event summary network meets the training completion condition. For example, training iterations reach 5 thousand times. The rate of change of the model parameters of the semantic recognition event summary network is less than 0.05. In this embodiment, setting the training completion condition can accelerate the model convergence speed.

In this embodiment, when the model parameter changes little and is smaller than the predetermined threshold, it is determined that the loss value of the semantic recognition event summary network converges, and the semantic recognition event summary model training is completed.

In some optional implementations of the disclosure, in response to the semantic recognition event summary network not satisfying the training completion condition, model parameters of the semantic recognition event summary network are adjusted such that a penalty function of the semantic recognition event summary network converges, continuing to perform steps 403-404 described above.

The embodiment of the disclosure provides a training method for a semantic recognition event summary model, firstly, a sample set is obtained, the sample set at least comprises one sample, and the sample comprises: text semantic features, features of merchandise order information corresponding to sample text semantic features; secondly, acquiring a pre-constructed semantic identification event summary network, wherein the semantic identification event summary network is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories; again, selecting a sample from the sample set; inputting the sample into a semantic recognition event summary network from time to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary network; and finally, responding to the semantic recognition event summary network to meet the training completion condition, and taking the semantic recognition event summary network as a semantic recognition event summary model. Therefore, the semantic recognition event summary network is trained through the selected sample, a trained semantic recognition event summary model is obtained, and reliability of the semantic recognition event summary model is improved.

In some optional implementations of the disclosure, the semantic recognition event summary network includes: the system comprises a text semantic feature network and a discrete feature network, wherein the text semantic feature network is used for representing the corresponding relation between text semantic features and various event summaries in preset categories; the discrete feature network is used for representing the corresponding relation between the features of commodity order information and various event summaries in the preset category.

In the optional implementation manner, in the training process of the semantic recognition event summary network, a text semantic feature network and a discrete feature network are trained jointly, the output of the semantic recognition event summary network is the probability of each event summary, and the loss is calculated between the output of the semantic recognition event summary network and the event summary selected by customer service. And the reliability of the optimal model training is ensured by the combined training on a large amount of historical data.

In this alternative implementation, the text semantic feature network is a neural network that processes text semantic features. The discrete feature network is a neural network for processing commodities and orders, the text semantic feature network and the discrete feature network are fused, and the semantic recognition event summary network can be trained jointly by using the same loss function.

The semantic recognition event summary network provided by the alternative implementation mode comprises the following components: the system comprises a text semantic feature network and a discrete feature network, wherein the text semantic feature network is used for representing the corresponding relation between text semantic features and various event summaries in preset categories. The discrete feature network is used for representing the corresponding relation between the features of commodity order information and various event summaries in preset categories, so that the text semantic feature network and the discrete feature network can be trained simultaneously through the same loss function in the process of training the semantic recognition event summary network, the semantic recognition event summary network can analyze and recognize various features simultaneously, and the probability of accurate event summaries is given.

With further reference to fig. 5, as an implementation of the method shown in the foregoing figures, the present disclosure provides an embodiment of an event summary generating apparatus, which corresponds to the method embodiment shown in fig. 2, and which is particularly applicable to various electronic devices.

As shown in fig. 5, an embodiment of the present disclosure provides an event summary generating apparatus 500, the apparatus 500 including: an information acquisition unit 501, a detection unit 502, a recording acquisition unit 503, and a result acquisition unit 504. The information obtaining unit 501 may be configured to obtain, when a customer service communicates with a customer, commodity order information related to the communication. The detection unit 502 may be configured to detect whether there is a recommendation result of the event summary based on the commodity order information. The recording acquisition unit 503 may be configured to acquire recording information of a call in response to detecting a recommendation result without an event summary. The result obtaining unit 504 may be configured to obtain a recommendation result of the event summary based on the commodity order information, the recording information and a pre-trained semantic recognition event summary model, where the semantic recognition event summary model is used to characterize a correspondence between text semantic features and features of the commodity order information and various event summaries in a preset category.

In the event summary generating apparatus 500 of the present embodiment, the specific processes of the information acquiring unit 501, the detecting unit 502, the recording acquiring unit 503, the result obtaining unit 504 and the technical effects thereof may refer to steps 201, 202, 203 and 204 in the corresponding embodiment of fig. 2, respectively.

In some embodiments, the commodity order information includes: customer information, merchant information, merchandise information, and order information; the detection unit 502 is further configured to: determining characteristics of customer information, merchant information, commodity information and order information; and detecting whether the recommended result of the event summary exists or not based on the consistent characteristics in the client information, the merchant information, the commodity information and the order information and a preset rule system.

In some embodiments, the above-described result obtaining unit 504 is further configured to: based on the recording information, obtaining text semantic features; extracting commodity order features based on commodity order information; inputting text semantic features and commodity order features into a pre-trained semantic recognition event summary model to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary model; and obtaining a recommendation result of the event summary based on the probabilities of the event summaries.

In some embodiments, the above-described result obtaining unit 504 is further configured to: and ordering the probabilities of all event summaries in the event summaries of the preset types in a descending order to obtain the probability of the pre-set bit, and taking the event summary corresponding to the probability of the pre-set bit as a recommendation result.

In some embodiments, the above-described result obtaining unit 504 is further configured to: determining thresholds of various event summaries based on historical data of the various event summaries; based on the probability of each event summary, detecting whether each event summary meets the requirement of each threshold; and responding to the requirement that the event summary meets the threshold value, and taking the event summary as a recommendation result.

In the event summary generating device provided in the embodiment of the present disclosure, first, when a customer service communicates with a customer, the information acquiring unit 501 acquires commodity order information related to the communication; next, the detection unit 502 detects whether or not there is a recommendation result of the event summary based on the commodity order information; again, the recording acquisition unit 503 acquires recording information of the call in response to detecting the recommended result without the event summary; finally, the result obtaining unit 504 obtains a recommendation result of the event summary based on the commodity order information, the recording information and the pre-trained semantic identification event summary model, where the semantic identification event summary model is used to characterize the corresponding relationship between the text semantic feature and the feature of the commodity order information and various event summaries in the preset category. Therefore, when the recommendation result of the event summary cannot be detected through commodity order information, the recommendation result of the event summary is obtained based on the call recording information, the summary log, the commodity order information and the pre-trained semantic recognition event summary model, and the semantic recognition event summary model is introduced to analyze the event summary on the basis that the recommendation result cannot be obtained through the commodity order information, so that a large amount of data distribution effects are achieved; by introducing the semantic recognition event summary model, the barrier for predicting the event summary only from customer service and customer call content is broken, and the accuracy of event summary prediction is improved.

With further reference to fig. 6, as an implementation of the method illustrated in the foregoing figures, the present disclosure provides an embodiment of a semantic recognition event summary model training apparatus, which corresponds to the method embodiment illustrated in fig. 4, and which is particularly applicable in various electronic devices.

As shown in fig. 6, an embodiment of the present disclosure provides a semantic recognition event summary model training apparatus 600, the apparatus 600 comprising: a sample acquisition unit 601, a network acquisition unit 602, a selection unit 603, an input unit 604, and a model obtaining unit 605. The sample acquiring unit 601 may be configured to acquire a sample set, where the sample set includes: at least one sample, each sample comprising: sample text semantic features, and features of merchandise order information corresponding to the sample text semantic features. The network obtaining unit 602 may be configured to obtain a semantic recognition event summary network, where the semantic recognition event summary network is used to characterize a correspondence between text semantic features, features of commodity order information, and various event summaries in a preset category. The selection unit 603 may be configured to select samples from a set of samples. The input unit 604 may be configured to input the sample into the semantic recognition event summary network, so as to obtain probabilities of various event summaries in a preset category output by the semantic recognition event summary network. The model obtaining unit 605 may be configured to obtain the semantic recognition event summary model corresponding to the semantic recognition event summary network in response to the semantic recognition event summary network satisfying the training completion condition.

In this embodiment, the semantic recognition event summary model is obtained by training using a semantic recognition event summary model training device.

In this embodiment, the specific processes of the sample acquiring unit 601, the network acquiring unit 602, the selecting unit 603, the input unit 604, and the model obtaining unit 605 and the technical effects thereof may refer to the steps 401, 402, 403, 404, and 405 in the corresponding embodiment of fig. 4, respectively.

Referring now to fig. 7, a schematic diagram of an electronic device 700 suitable for use in implementing embodiments of the present disclosure is shown.

As shown in fig. 7, the electronic device 700 may include a processing means (e.g., a central processor, a graphics processor, etc.) 701, which may 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 means 708 into a Random Access Memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 are also stored. The processing device 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 bus 704.

In general, the following devices may be connected to the I/O interface 705: input devices 706 including, for example, a touch screen, touchpad, keyboard, mouse, etc.; output device 707 including, for example, a Liquid Crystal Display (LCD), a speaker, a vibrator, and the like; storage 708 including, for example, magnetic tape, hard disk, etc.; and a communication device 709. The communication means 709 may allow the electronic device 700 to communicate wirelessly or by wire with other devices to exchange data. While fig. 7 shows an electronic device 700 having various means, it is to be understood that not all of the illustrated means are required to be implemented or provided. More or fewer devices may be implemented or provided instead. Each block shown in fig. 7 may represent one device or a plurality of devices as needed.

In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication device 709, or installed from storage 708, or installed from ROM 702. The above-described functions defined in the methods of the embodiments of the present disclosure are performed when the computer program is executed by the processing device 701.

It should be noted that the computer readable medium of the embodiments of the present disclosure may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In an embodiment of the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. Whereas in embodiments of the present disclosure, the computer-readable signal medium may comprise a data signal propagated in baseband or as part of a carrier wave, with computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: electrical wires, fiber optic cables, RF (Radio Frequency), and the like, or any suitable combination thereof.

The computer readable medium may be contained in the server; or may exist alone without being assembled into the server. The computer readable medium carries one or more programs which, when executed by the server, cause the server to: when customer service communicates with a customer, acquiring commodity order information related to communication; detecting whether a recommended result of the event summary exists or not based on commodity order information; acquiring recording information of a call in response to detecting a recommendation result without an event summary; based on commodity order information, recording information and a pre-trained semantic recognition event summary model, a recommendation result of the event summary is obtained, and the semantic recognition event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories.

Computer program code for carrying out operations of embodiments of the present disclosure may be written in one or more programming languages, including an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units involved in the embodiments described in the present disclosure may be implemented by means of software, or may be implemented by means of hardware. The described units may also be provided in a processor, for example, described as: a processor comprises an information acquisition unit, a detection unit, a recording acquisition unit and a result obtaining unit. The names of these units do not constitute a limitation on the unit itself in some cases, and for example, the information acquisition unit may also be described as a "unit configured to acquire commodity order information related to a call when a customer service calls with a customer".

The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above technical features, but encompasses other technical features formed by any combination of the above technical features or their equivalents without departing from the spirit of the invention. Such as the above-described features, are mutually substituted with (but not limited to) the features having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method of event summary generation, the method comprising:

When customer service communicates with a customer, acquiring commodity order information related to the communication;

Detecting whether a recommended result of the event summary exists or not based on the commodity order information;

acquiring recording information of the call in response to detecting that the recommended result of the event summary is not available;

And obtaining a recommendation result of the event summary based on the commodity order information, the recording information and a pre-trained semantic identification event summary model, wherein the semantic identification event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories.

2. The method of claim 1, wherein the merchandise order information comprises: customer information, merchant information, merchandise information, and order information; the detecting whether the recommended result of the event summary exists or not based on the commodity order information comprises:

determining characteristics of the customer information, the merchant information, the merchandise information, and the order information;

And detecting whether a recommendation result of the event summary exists or not based on the consistent characteristics in the client information, the merchant information, the commodity information and the order information and a preset rule system.

3. The method of claim 1, wherein the obtaining a recommendation for the event summary based on the merchandise order information, the sound recording information, and a pre-trained semantic recognition event summary model comprises:

based on the recording information, obtaining text semantic features;

extracting commodity order features based on the commodity order information;

Inputting the text semantic features and the commodity order features into a pre-trained semantic identification event summary model to obtain probabilities of various event summaries in preset categories output by the semantic identification event summary model;

and obtaining a recommendation result of the event summary based on the probability of various event summaries.

4. The method of claim 3, wherein the deriving the recommendation for the event summary based on probabilities for various types of event summaries comprises:

And ordering the probabilities of all event summaries in the event summaries of the preset types in a descending order to obtain the probability of the pre-set bit, and taking the event summary corresponding to the probability of the pre-set bit as a recommendation result.

5. The method of claim 3, wherein the deriving the recommendation for the event summary based on probabilities for various types of event summaries comprises:

determining thresholds of various event summaries based on historical data of the various event summaries;

based on the probability of each event summary, detecting whether each event summary meets the requirement of each threshold;

and responding to the requirement that the event summary meets the threshold value, and taking the event summary as a recommendation result.

6. A semantic recognition event summary model training method, the method comprising:

Obtaining a sample set, the sample set comprising: at least one sample, each sample comprising: sample text semantic features, and features of merchandise order information corresponding to the sample text semantic features;

Acquiring a semantic recognition event summary network, wherein the semantic recognition event summary network is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories;

The following training steps are performed:

Selecting a sample from the sample set;

inputting the sample into the semantic recognition event summary network to obtain probabilities of various event summaries in preset categories output by the semantic recognition event summary network;

And responding to the semantic recognition event summary network to meet the training completion condition, and obtaining a semantic recognition event summary model corresponding to the semantic recognition event summary network.

7. The method of claim 6, the semantic recognition event summary network comprising: the system comprises a text semantic feature network and a discrete feature network, wherein the text semantic feature network is used for representing the corresponding relation between text semantic features and various event summaries in a preset category; the discrete feature network is used for representing the corresponding relation between the features of commodity order information and various event summaries in the preset category.

8. An event summary generation apparatus, the apparatus comprising:

an information acquisition unit configured to acquire commodity order information related to a call when a customer service calls with a customer;

a detection unit configured to detect whether a recommended result of an event summary is present or not based on the commodity order information;

A recording acquisition unit configured to acquire recording information of the call in response to detecting that the recommended result of the event summary is not present;

The result obtaining unit is configured to obtain a recommended result of the event summary based on the commodity order information, the recording information and a pre-trained semantic identification event summary model, wherein the semantic identification event summary model is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories.

9. A semantic recognition event summary model training apparatus, the apparatus comprising:

A sample acquisition unit configured to acquire a sample set including: at least one sample, each sample comprising: sample text semantic features, and features of merchandise order information corresponding to the sample text semantic features;

The network acquisition unit is configured to acquire a semantic recognition event summary network, wherein the semantic recognition event summary network is used for representing the corresponding relation between text semantic features and features of commodity order information and various event summaries in preset categories;

a selecting unit configured to select a sample from the sample set;

the input unit is configured to input the sample into the semantic recognition event summary network to obtain the probability of various event summaries in a preset category output by the semantic recognition event summary network;

the model obtaining unit is configured to obtain a semantic recognition event summary model corresponding to the semantic recognition event summary network in response to the semantic recognition event summary network meeting training completion conditions.

10. An electronic device, comprising:

one or more processors;

A storage device having one or more programs stored thereon;

when executed by the one or more processors, causes the one or more processors to implement the method of any of claims 1-7.

11. A computer readable medium having stored thereon a computer program, wherein the program when executed by a processor implements the method of any of claims 1-7.