Event processing method, device and equipment

Abstract

The present application provides an event processing method, device and equipment, the method comprising: obtaining at least one trigger word according to original text data; processing the at least one trigger word based on a trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type; determining a target event according to the event type and the event element; and processing the target event according to the statistical information of the target event. Through the technical solution of the present application, the target neural network is used to identify and extract event types and event elements, effectively improving the utilization rate of context.

Description

Technical Field

[0001] The present application relates to the field of Internet, and in particular to an event processing method, device and equipment. Background Art

[0002] With the changes in business and technology in the customer service industry, the customer service industry has attached more and more importance to data analysis, hoping to deeply explore business value, consumer behavior, market information, and work efficiency through data analysis to achieve convenient and fast intelligence. In the e-commerce application scenario, the review analysis system is an independent system that can operate across multiple e-commerce platforms. By capturing the customer service data of the e-commerce platform and conducting in-depth analysis of the customer service data of the e-commerce platform, it can explore user concerns, thereby improving products, increasing repurchase rates, and increasing sales.

[0003] However, in order to achieve in-depth analysis of customer service data, in relevant technologies, analysts are usually required to conduct in-depth analysis of the customer service data of the e-commerce platform, that is, the participation of analysts is required. Obviously, since the amount of customer service data of the e-commerce platform is large, when analysts participate in the in-depth analysis of the customer service data of the e-commerce platform, the workload of the analysts is very large, and it is easy to have problems such as analysis errors. Summary of the invention

[0004] The present application provides an event processing method, the method comprising:

[0005] Acquire at least one trigger word according to the original text data;

[0006] Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0007] Determine a target event according to the event type and the event element;

[0008] The target event is processed according to the statistical information of the target event.

[0009] The present application provides an event processing method, the method comprising:

[0010] Acquire an event type corresponding to the original text data and an event element corresponding to the event type;

[0011] Determine a target event according to the event type and the event element;

[0012] The target event is processed according to the statistical information of the target event.

[0013] The present application provides an event processing device, the device comprising:

[0014] An acquisition module, used for acquiring at least one trigger word according to the original text data;

[0015] A processing module, configured to process the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0016] A determination module, used for determining a target event according to the event type and the event element;

[0017] The processing module is further used to process the target event according to the statistical information of the target event.

[0018] The present application provides an event processing device, comprising:

[0019] A processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a plurality of computer instructions, and when the processor executes the computer instructions, the processor performs the following processing:

[0020] Acquire at least one trigger word according to the original text data;

[0021] Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0022] Determine a target event according to the event type and the event element;

[0023] The target event is processed according to the statistical information of the target event.

[0024] Based on the above technical solution, in the embodiment of the present application, the event type corresponding to the original text data and the event element corresponding to the event type can be obtained according to the target neural network, the target event can be determined according to the event type and event element, and the target event can be processed according to the statistical information of the target event, so that the target neural network is used to identify and extract the event type and event element, which effectively improves the utilization rate of the context, helps to solve the ambiguous problem, and upgrades the warning from the word and phrase granularity to the warning based on the "event" level, greatly enriching the information dimension of the warning result, improving the readability, and being able to present the information to the user in a relatively complete and comprehensive manner. Moreover, there is no need for analysts to conduct in-depth analysis of the original text data of the e-commerce platform, which solves the problems of heavy workload of analysts and easy analysis errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings of the embodiments of the present application.

[0026] Figure 1 It is a flowchart of an event processing method in one embodiment of the present application;

[0027] Figure 2 is a flowchart of an event processing method in another embodiment of the present application;

[0028] Figure 3 is a flowchart of an event processing method in another embodiment of the present application;

[0029] Figure 4 is a flowchart of an event processing method in another embodiment of the present application;

[0030] Figure 5 is a schematic diagram of the structure of an event processing device in one embodiment of the present application;

[0031] Figure 6 It is a structural diagram of an event processing device in one embodiment of the present application. DETAILED DESCRIPTION

[0032] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, rather than limiting the present application. The singular forms of "a", "said" and "the" used in the present application and claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0033] It should be understood that, although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, in addition, the word "if" used may be interpreted as "at..." or "when..." or "in response to determination".

[0034] An event processing method is proposed in the embodiment of the present application, which can be applied to any application scenario, such as an e-commerce application scenario, without limitation. The e-commerce application scenario is taken as an example below, and the event processing method is used to realize the tracking and early warning of real-time events in the e-commerce field. It can perform multi-dimensional structured expression of the problems / opportunities reflected by consumers at the granularity of "event", and track and warn the events throughout their life cycle.

[0035] See also Figure 1 FIG. 1 is a flow chart of the event processing method, which may include:

[0036] Step 101: Obtain at least one trigger word according to original text data.

[0037] For example, the original text data may be obtained, and the original text data may be data of a consumer for a certain product, such as the original text data is customer service data of an e-commerce platform or evaluation data of an e-commerce platform, and the type of the original text data is not limited. For example, the original text data may be: "I complained that the xx store sold fake goods. I bought a pair of sports shoes A from the store a few days ago, but found that they were high-quality imitations of the yy brand shoes."

[0038] After obtaining the original text data, the original text data can be divided into at least one trigger word. For example, a word segmenter is used to divide the original text data into trigger words. There is no restriction on the division method, as long as the original text data can be divided into at least one trigger word. The following trigger words can be obtained for the above original text data: complaint, xx store, sale, fake, shoes, A-type sports shoes, high imitation, yy brand, etc.

[0039] In one possible implementation, a customer service log of a consumer-manual customer service can be obtained from a customer service system, or a customer service log of a consumer-customer service robot can be obtained from a customer service system. The customer service log is raw text data. Of course, the above are just two examples of obtaining raw text data and there is no limitation to this.

[0040] After obtaining the original text data, the original text data may be preprocessed to obtain at least one trigger word. For example, the original text data may be cleaned and / or filtered. In the data cleaning and filtering process, redundant information in the original text data may be removed, without limitation, such as removing redundant information such as numeric strings, swear words, and URLs (Uniform Resource Locators).

[0041] After the original text data is processed by data cleaning and / or data filtering, the processed original text data is processed by word segmentation, stop word removal, synonym normalization, etc. to obtain at least one trigger word.

[0042] Step 102: Process at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type.

[0043] Exemplarily, the at least one trigger word can be input into a trained target neural network so that the target neural network determines an event type trigger word based on the at least one trigger word, and determines the event type corresponding to the original text data based on the event type trigger word, and determines an event element trigger word based on the at least one trigger word, and determines the event element corresponding to the event type based on the event element trigger word.

[0044] Among them, the event type trigger word is a trigger word used to determine the event type among the at least one trigger word, and the event element trigger word is a trigger word used to determine the event element among the at least one trigger word.

[0045] Determine an event element trigger word based on the at least one trigger word, and determine an event element corresponding to the event type based on the event element trigger word, including but not limited to: determine an element role corresponding to the event type based on the target neural network, determine an event element trigger word corresponding to the element role based on at least one trigger word, and determine an event element corresponding to the element role based on the event element trigger word.

[0046] Exemplarily, in step 102, concepts such as event type, event element, and element role are involved. These concepts are explained below in conjunction with specific application scenarios. An event is a structured representation of the participating subject, the object of action, and the relationship and state between the two in the original text data. In other words, an event is an objective fact that describes the relationship between the subject and the object of action. Event extraction is to extract this subject-object-relationship combination from the original text data. The event type is the field to which the event belongs, the element role is the entity type name of the event, and the event element is the entity enumeration and description of the element role, that is, the value of the element role.

[0047] For example, event templates corresponding to multiple event types can be pre-configured. The event templates can be shown in Table 1 or Table 2. Table 1 is the event template corresponding to the event type "fake complaint", and Table 2 is the event template corresponding to the event type "return and refund". Of course, Tables 1 and 2 are just examples and there is no restriction on this event type. For different event types, event templates corresponding to the event type can be configured.

[0048] Table 1

[0049] Event Type Fake goods complaints time Category Shop commodity

[0050] Table 2

[0051] Event Type Return and Refund time Category Shop commodity

[0052] In Table 1 and Table 2, time, category, store, product, etc. represent element roles. Further, after adding event elements to the event template, see Table 3 and Table 4.

[0053] Table 3

[0054] Event Type Fake goods complaints time x month x day Category shoe Shop xx shop commodity A-type sports shoes

[0055] Table 4

[0056] Event Type Return and Refund time x month x day Category dress Shop xx shop commodity C-style dress

[0057] As shown in Table 3, the event elements corresponding to the event type "fake goods complaint" are recorded, "shoes" represents the event element corresponding to the element role "category", "xx store" represents the event element corresponding to the element role "store", and "A style sports shoes" represents the event element corresponding to the element role "commodity".

[0058] As shown in Table 4, the event elements corresponding to the event type "return and refund" are recorded, "dress" represents the event element corresponding to the element role "category", "xx store" represents the event element corresponding to the element role "store", and "C dress" represents the event element corresponding to the element role "product".

[0059] In step 102, in fact, at least one trigger word corresponding to the original text data is processed based on the target neural network to obtain the event type (such as counterfeit complaints, etc.) and each event element (such as shoes, xx store, A style sports shoes, etc.). This process is explained in combination with specific application scenarios below.

[0060] For example, based on the trigger words corresponding to the original text data, such as complaint, xx store, sale, fake, shoes, model A sports shoes, high imitation, yy brand, etc., these trigger words can be input into the trained target neural network so that the target neural network can process these trigger words without any restriction on this processing process.

[0061] After processing these trigger words, the target neural network can determine the event type trigger words based on these trigger words. The event type trigger words are trigger words used to determine the event type, such as the event type trigger words are "fake goods" and "complaint". Then, the target neural network can determine the event type based on the event type trigger words, such as determining the event type as "fake goods complaint" based on "fake goods" and "complaint".

[0062] Furthermore, after obtaining the event type, the target neural network can determine the element role corresponding to the event type, such as determining the element role corresponding to the event type based on the event template corresponding to the event type "fake goods complaint" shown in Table 1, such as the element role is time, category, store, commodity, etc.

[0063] For example, after the target neural network processes the above trigger words, it can also determine the event element trigger words corresponding to each element role based on these trigger words. The event element trigger words are trigger words used to determine event elements. The target neural network can determine the event element corresponding to the element role based on the event element trigger words corresponding to the element role. For example, the event element trigger word corresponding to the element role "category" is "shoes", and then the event element corresponding to the element role "category" can be determined to be "shoes". The event element trigger word corresponding to the element role "store" is "xx store", and then the event element corresponding to the element role "store" can be determined to be "xx store". The event element trigger word corresponding to the element role "commodity" is "A style sports shoes", and then the event element corresponding to the element role "commodity" can be determined to be "A style sports shoes".

[0064] The element role corresponding to the event type also includes "time". Therefore, the occurrence time of the original text data (indicating when the original text data was generated) can also be input into the target neural network. The target neural network determines the event element corresponding to the element role "time" based on the occurrence time of the original text data. For example, the event element can be the occurrence time of the original text data, such as x month x day.

[0065] To summarize, after the trigger words corresponding to the original text data are input into the target neural network, the target neural network can output the event type corresponding to the original text data and the event elements corresponding to the event type.

[0066] Step 103: determine the target event according to the event type and the event element.

[0067] After obtaining the event type and event element, the event type and event elements can be combined into a target event. For example, after obtaining the event type "fake complaint", event element "shoes", event element "xx store", event element "type A sports shoes", and event element "x month x day", the target event is determined based on these event types and event elements. See Table 3, which is an example of a target event and is recorded as target event A. The event type of target event A is "fake complaint", and the event elements of target event A include: event element "shoes" corresponding to the element role "category", event element "xx store" corresponding to the element role "store", event element "type A sports shoes" corresponding to the element role "commodity", and event element "x month x day" corresponding to the element role "time".

[0068] Step 104: Process the target event according to the statistical information of the target event.

[0069] In a possible implementation, the alarm score of the target event can be determined based on the statistical information of the target event, and the statistical information indicates that the more the statistics of the target event are, the greater the alarm score is. If the alarm score is greater than the threshold, the alarm information of the target event is sent to the target device.

[0070] Exemplarily, determining the alarm score of the target event based on the statistical information of the target event may include, but is not limited to: determining the alarm score of the target event based on the statistical information of the target event and the importance of the target event; wherein the importance indicates that the more important the target event is, the larger the alarm score is.

[0071] To sum up, for the target event, the statistical information of the target event can be determined, and the statistical information represents the statistical value of the target event. For example, based on the original text data 1, the target event is determined to be target event A, and the statistics of target event A are updated to 1; based on the original text data 2, the target event is determined to be target event A, and the statistics of target event A are updated to 2; based on the original text data 3, the target event is determined to be target event B, and the statistics of target event B are updated to 1; based on the original text data 4, the target event is determined to be target event A, and the statistics of target event A are updated to 3, and so on.

[0072] For the target event, the importance of the target event can also be determined. The importance can be unimportant, generally important, very important, etc. There is no restriction on the level of importance and it can be configured based on experience. In addition, the mapping relationship between event type and importance can be pre-configured, such as the mapping relationship between return and refund and generally important, the mapping relationship between counterfeit complaints and very important, etc. There is no restriction on this mapping relationship and it can be configured based on experience. In summary, based on the event type of the target event, the importance of the target event can be obtained by querying the above mapping relationship through the event type.

[0073] After obtaining the statistical information of the target event and the importance of the target event, the alarm score of the target event can be determined according to the statistical information of the target event and the importance of the target event; wherein, when the importance indicates that the target event is more important, the alarm score is larger, and when the statistical information indicates that the statistical quantity of the target event is larger, the alarm score is larger. For example, the statistical information of the target event and the importance of the target event are normalized to obtain the statistical score value of the target event and the importance score value of the target event, and the statistical score value and the importance score value are weighted to obtain the alarm score of the target event.

[0074] For example, when normalizing the statistical information of the target event, the larger the target event's statistic, the larger the target event's statistical score, and the smaller the target event's statistic, the smaller the target event's statistical score. For example, when the target event A's statistic is 1-10, the statistical score is 10, and when the target event A's statistic is 11-20, the statistical score is 20, and so on. There is no restriction on this normalization process.

[0075] When normalizing the importance of the target event, the more important the target event is, the greater the importance score of the target event is, and the less important the target event is, the smaller the importance score of the target event is. For example, when the importance of target event A is unimportant, the importance score is 30, when the importance of target event A is moderately important, the importance score is 60, and when the importance of target event A is very important, the importance score is 90. There is no restriction on this normalization process.

[0076] Based on the statistical score value and the importance score value, the alarm score of the target event can be obtained by the following formula: S = S1*W1+S2*W2, S1 represents the statistical score value, W1 represents the weight coefficient of the statistical score value, S2 represents the importance score value, and W2 represents the weight coefficient of the importance score value. The values ​​of W1 and W2 can be configured based on experience and are not limited. For example, W1 is a value between 0 and 1, W2 is a value between 0 and 1, and the sum of W1 and W2 can be 1. In practical applications, when the statistical score value is emphasized, W1 can be greater than W2, and when the importance score value is emphasized, W2 can be greater than W1.

[0077] To sum up, the alarm score of the target event can be obtained. If the alarm score is greater than the threshold (configured based on experience, there is no limit on this threshold), the alarm information of the target event will be sent to the target device (for example, when the target event includes "xx store", the target device is the device corresponding to the merchant of "xx store").

[0078] In one example, the above execution order is just an example given for the convenience of description. In practical applications, the execution order between the steps can also be changed, and there is no limitation on this execution order. Moreover, in other embodiments, the steps of the corresponding method are not necessarily executed in the order shown and described in this specification, and the steps included in the method may be more or less than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; multiple steps described in this specification may also be combined into a single step for description in other embodiments.

[0079] Based on the above technical solution, in the embodiment of the present application, the event type corresponding to the original text data and the event element corresponding to the event type can be obtained according to the target neural network, the target event can be determined according to the event type and event element, and the target event can be processed according to the statistical information of the target event, so that the target neural network is used to identify and extract the event type and event element, which effectively improves the utilization rate of the context, helps to solve the ambiguous problem, and upgrades the warning from the word and phrase granularity to the warning based on the "event" level, greatly enriching the information dimension of the warning result, improving the readability, and being able to present the information to the user in a relatively complete and comprehensive manner. Moreover, there is no need for analysts to conduct in-depth analysis of the original text data of the e-commerce platform, which solves the problems of heavy workload of analysts and easy analysis errors.

[0080] In one possible implementation, an initial neural network may be obtained and trained to obtain a trained target neural network. Step 102 may be performed based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type. The following describes the training process of the initial neural network and the use process of the target neural network.

[0081] During the training process of the initial neural network, a large amount of training text data can be obtained. For each training text data, at least one trigger word is obtained according to the training text data, and the initial neural network is trained based on the at least one trigger word corresponding to the training text data to obtain a trained target neural network; the target neural network includes a mapping relationship between event type trigger words and event types, and a mapping relationship between event element trigger words and event elements; the event type trigger word is a trigger word in at least one trigger word used to determine the event type; the event element trigger word is a trigger word in at least one trigger word used to determine the event element.

[0082] For example, for each trigger word corresponding to the training text data, the trigger word may have calibration information, and the calibration information indicates that the trigger word is an event type trigger word, or the trigger word is an event element trigger word, or the trigger word is neither an event type trigger word nor an event element trigger word. If the trigger word is an event type trigger word, the calibration information also indicates the event type corresponding to the trigger word, and if the trigger word is an event element trigger word, the calibration information also indicates the event element corresponding to the trigger word.

[0083] Then, each trigger word corresponding to the training text data and the calibration information of each trigger word are input into the initial neural network, so as to train the initial neural network using these trigger words and calibration information. There is no restriction on this training process, and various types of neural network training methods can be used. After the training process is completed, a trained target neural network can be obtained, and the target neural network can include a mapping relationship between event type trigger words and event types, and a mapping relationship between event element trigger words and event elements.

[0084] During the initial neural network training process, for multiple trigger words corresponding to the training text data, the following situation exists: a combination of N (a positive integer greater than or equal to 1) trigger words can correspond to the same event type. For example, the calibration information of the N trigger words all indicates that the trigger word is an event type trigger word, and the calibration information of these N trigger words indicates that these N trigger words correspond to the same event type.

[0085] During the initial neural network training process, for multiple trigger words corresponding to the training text data, the following situation exists: a combination of M (a positive integer greater than or equal to 1) trigger words can correspond to the same event element. For example, the calibration information of the M trigger words all indicates that the trigger word is an event element trigger word, and the calibration information of these M trigger words indicates that these M trigger words correspond to the same event element.

[0086] During the initial neural network training process, for a trigger word corresponding to the training text data, if the trigger word is an event element trigger word, the calibration information of the trigger word may also indicate an element role.

[0087] In summary, a trained target neural network can be obtained, and this process will not be described in detail. After obtaining the trained target neural network, for step 101 and step 102, the original text data can be obtained, and at least one trigger word can be obtained based on the original text data, and these trigger words are input into the target neural network, and the target neural network processes these trigger words to obtain event type trigger words. Since the target neural network includes a mapping relationship between event type trigger words and event types, the target neural network can determine the event type corresponding to the original text data based on the event type trigger word, and this will not be described in detail.

[0088] After obtaining the event type corresponding to the original text data, the target neural network can also determine at least one element role corresponding to the event type. For each element role, when the target neural network processes these trigger words, it can obtain the event element trigger word corresponding to the element role. Since the target neural network includes a mapping relationship between the event element trigger word and the event element, the target neural network can determine the event element corresponding to the element role based on the event element trigger word, which will not be described in detail.

[0089] To sum up, the target neural network can obtain the event type corresponding to the original text data and the event elements corresponding to each element role (i.e., the event elements corresponding to the event type). At this point, the target neural network can output the event type corresponding to the original text data and the event elements corresponding to the event type.

[0090] In the above embodiment, the structure of the target neural network can adopt the network structure of GRU (Gate Recurrent Unit) and CRF (Conditional random field). Of course, the network structure of GRU and CRF is just an example, and there is no restriction on the structure of the target neural network. GRU is a kind of recurrent neural network, which is proposed to solve the long-term memory dependency problem in back propagation. CRF is a conditional probability distribution model of another set of output sequences under the condition of a set of input sequences.

[0091] To sum up, in this embodiment, a bidirectional GRU network can be used, and an external CRF model can be used to solve the maximum probability path distribution, so as to jointly extract event types, event elements, and element roles. The context environment can be used more efficiently, and the ambiguity problem of event type and event element extraction can be effectively solved. The trigger words and the context in which the trigger words are located can be analyzed, and the event types, event elements, and element roles can be effectively combined. The context environment can be shared with each other to avoid being separated from each other.

[0092] In a possible implementation manner, for step 103, based on the event type and the event element, see Figure 2 As shown, the target event can be determined by the following steps:

[0093] Step 1031: Determine the normalization element corresponding to the event element based on the type of the event element.

[0094] Exemplarily, the type of the event element can be a commodity domain type or a general domain type. The commodity domain type indicates that the event element is a commodity, and the general domain type indicates that the event element is not a commodity. For example, for the above event element "shoes", the type of the event element "shoes" is a commodity domain type. For the above event element "type A sports shoes", the type of the event element "type A sports shoes" is a commodity domain type. For the above event element "xx store", the type of the event element "xx store" is a general domain type.

[0095] Exemplarily, the process of determining the normalized element corresponding to the event element is actually the process of entity unification. The task of entity unification is to normalize different expressions of the same type of things (each expression corresponds to an event element) to the same expression (the normalized expression corresponds to the normalized element). The symbols used are as follows: event element-E(i,j), normalized element-E^(j), j represents the jth normalized element, i represents the i-th expression of the jth normalized element, that is, the i-th event element.

[0096] For example, "Style A sneakers" can be "retro men's sneakers", and the expression of "retro men's sneakers" can be: "retro men's sneakers", "2020 new ethnic style retro men's shoes", "new retro low-top men's sneakers", "ethnic style retro sports style men's shoes", and the above expressions need to be normalized to "retro men's sneakers". For example, in step 102, if the event element output by the target neural network is "retro men's sneakers", then the normalized element corresponding to the event element is "retro men's sneakers", if the event element output by the target neural network is "2020 new ethnic style retro men's shoes", then the normalized element corresponding to the event element is "retro men's sneakers", if the event element output by the target neural network is "new retro low-top men's sneakers", then the normalized element corresponding to the event element is "retro men's sneakers".

[0097] In order to normalize different expressions of the same type of things (i.e., different event elements) to the same expression (i.e., normalized element), the enumeration space of event elements can be divided into a product domain and a general domain. The product domain is the enumeration space of all product names, and the general domain is the enumeration space excluding products.

[0098] In the above application scenario, the normalization element corresponding to the event element may include but is not limited to:

[0099] Method 1: If the type of the event element is a product domain type, the normalization element corresponding to the event element can be determined based on the product knowledge graph; wherein the product knowledge graph can be pre-configured, and the product knowledge graph can include a mapping relationship between the event element and the normalization element.

[0100] For example, a product knowledge graph can be pre-configured, and the product knowledge graph is used to record the mapping relationship between event elements and normalized elements, and the content of the product knowledge graph is not limited. See Table 5 for an example of a product knowledge graph. Of course, this is just an example and is not limited to this.

[0101] Table 5

[0102] Event Elements Normalize elements 2020 new ethnic retro men's shoes Retro Men's Sneakers New retro low-top men's sneakers Retro Men's Sneakers Ethnic retro style sports style men's shoes Retro Men's Sneakers … …

[0103] Based on the above commodity knowledge graph, if the event element output by the target neural network is "retro men's sports shoes", since "retro men's sports shoes" already exists as a normalized element in the commodity knowledge graph, it is directly determined that the normalized element corresponding to the event element is "retro men's sports shoes". If the event element output by the target neural network is "2020 new ethnic retro men's shoes", by querying the commodity knowledge graph shown in Table 5, it is determined that the normalized element corresponding to the event element is "retro men's sports shoes".

[0104] To sum up, based on the product knowledge graph, for the event element -E(i,j) of the product domain type, the normalized element -E^(j) corresponding to the event element -E(i,j) can be queried. For example, for event elements such as "2020 new ethnic style retro men's shoes", "new retro low-top men's sports shoes", "ethnic style retro sports style men's shoes", etc., the corresponding normalized element can be queried as "retro men's sports shoes", which will not be repeated here.

[0105] Method 2: If the type of the event element is a general domain type, the target element entity corresponding to the event element can be determined from the element entity based on the similarity between the event element and the element entity; the normalized element corresponding to the target element entity is determined based on the generated element dictionary; wherein the element dictionary includes a mapping relationship between the normalized element and at least one element entity. Exemplarily, the process of generating the element dictionary may include but is not limited to: obtaining multiple element entities, clustering the multiple element entities based on a density clustering algorithm to obtain an element set, the element set including at least one element entity; generating the element dictionary based on the normalized element corresponding to the element set and at least one element entity included in the element set.

[0106] For example, for the entity unification of general domains, a combination of element dictionary mapping and semantic similarity calculation can be used. The process of generating the element dictionary can include: training text data--> word segmentation--> part-of-speech filtering, commodity filtering--> word vectorization--> density-based clustering (DBSCAN)--> generating an initial dictionary--> manual verification--> generating a complete element dictionary, and the element dictionary is in the form of: entity A-(entity A1, entity A2, ..., entity Ai). For the semantic similarity calculation process, it can include: for the candidate set of event elements, traverse each event element, first vectorize the event element, and then calculate the similarity with the element entity in the existing event. If the set threshold is met, the two are classified as the same entity.

[0107] The following describes the element dictionary mapping and semantic similarity calculation process in combination with specific application scenarios.

[0108] Acquire multiple training text data, perform word segmentation on these training text data to obtain multiple words, perform part-of-speech filtering (i.e., filter out words with specified parts of speech) and product filtering (i.e., filter out words of product domain type) on these words, and remove duplicate words (when multiple words are repeated, only one word is retained). In this way, the remaining words are used as element entities. At this point, multiple element entities can be obtained.

[0109] Each element entity is word vectorized to obtain the word vector of each element entity, that is, the element entity is represented by a word vector. Then, multiple element entities are clustered based on a density clustering algorithm (such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), etc.) to obtain an element set, which includes at least one element entity. When clustering multiple element entities, clustering is based on the word vector of each element entity, and there is no restriction on this clustering process. For example, element entity A1, element entity A2, ..., element entity Ai are clustered into element set A, element entity B1, element entity B2, ..., element entity Bi are clustered into element set B, and so on.

[0110] Then, an initial dictionary can be generated, which may include each element set and all its corresponding element entities, such as element set A (element entity A1, element entity A2, ..., element entity Ai), element set B (element entity B1, element entity B2, ..., element entity Bi), and so on.

[0111] Determine the normalized element corresponding to each element set, and there is no restriction on the determination method. For example, a manual verification method is adopted, and the user provides the normalized element corresponding to the element set. Assuming that all element entities in element set A (element entity A1, element entity A2, ..., element entity Ai) correspond to the same normalized element A, then the normalized element corresponding to element set A is normalized element A. Assuming that all element entities in element set B (element entity B1, element entity B2, ..., element entity Bi) correspond to the same normalized element B, then the normalized element corresponding to element set B is normalized element B.

[0112] Then, based on the normalized elements corresponding to the element set and the element entities included in the element set, an element dictionary can be generated. The element dictionary may include but is not limited to: normalized elements A-(element entity A1, element entity A2, ..., element entity Ai) corresponding to element set A, normalized elements B-(element entity B1, element entity B2, ..., element entity Bi) corresponding to element set B, and so on.

[0113] In step 102, the target neural network can output at least one event element, add these event elements to the candidate set of event elements, traverse each event element in the candidate set, and for the currently traversed event element, if the type of the event element is a general domain type, determine the similarity between the event element and the element entities in the existing events (such as the above-mentioned element entity A1, element entity A2,..., element entity Ai, element entity B1, element entity B2,..., element entity Bi, etc.).

[0114] If the similarity between the event element and a certain element entity (taking element entity A1 as an example) is greater than a set threshold (which can be configured based on experience), the two are classified as the same entity, that is, the event element belongs to element entity A1, that is, the target element entity corresponding to the event element is element entity A1.

[0115] When determining the similarity between the event element and the element entity in the existing event, the event element may be vectorized, and the element entity in the existing event is also the vectorized element entity.

[0116] After obtaining that the target element entity is element entity A1, the normalized element corresponding to the target element entity is determined based on the generated element dictionary. That is, the generated element dictionary is queried through element entity A1 to obtain the normalized element A corresponding to element entity A1, that is, the normalized element A corresponding to the event element is obtained.

[0117] Step 1032: Determine a target event according to the event type and the normalized element.

[0118] Exemplarily, after obtaining the event type and the normalized element (i.e., the normalized event element), the event type and the normalized element can be combined into a target event. For example, after obtaining the event type "fake complaint", the normalized element "shoes", the normalized element "xx store", the normalized element "style A sports shoes", and the normalized element "x month x day", the target event can be determined based on these event types and normalized elements.

[0119] In a possible implementation, it is possible to use Figure 3 The process shown in the figure implements the event handling method.

[0120] First, original text data is obtained. The original text data may be data of consumers regarding a certain product. For example, the original text data may be customer service data of an e-commerce platform or evaluation data of an e-commerce platform.

[0121] Then, the original text data is preprocessed to obtain at least one trigger word. For example, the original text data is cleaned and / or filtered, and the processed original text data is segmented, stop words are removed, synonyms are normalized, etc. to obtain at least one trigger word.

[0122] Then, event type recognition and event element recognition are performed. For example, in the process of event type recognition and event element recognition, at least one trigger word is processed based on the trained target neural network to obtain the event type corresponding to the original text data and the event element corresponding to the event type.

[0123] Then, the target event is determined according to the event type and the event element, and the state of the target event is tracked. For example, after determining the target event A, it can be determined whether the target event A exists in the existing target events. If it exists, the state of the target event A is updated, as shown in Table 6. If it does not exist, a target event A is created, that is, the target event A is used as the existing target event.

[0124] Table 6

[0125]

[0126] Exemplarily, judging whether there is a target event A in the existing target events may include: for each existing target event, determining the similarity between the target event A and the existing target event, if the similarity is greater than a preset similarity threshold, determining that the existing target event is the target event A, that is, the target event A exists in the existing target event, and updating the state of the target event A to the existing target event, as shown in Table 6. If the similarity between the target event A and all the existing target events is not greater than the preset similarity threshold, determining that the target event A does not exist in the existing target event, and a new record needs to be added.

[0127] In summary, after a target event is initially created, the new target event can be similarly calculated with the existing target event by streaming clustering. If the similarity meets a certain threshold, the new target event is classified as the existing target event, and the state of the existing target event is refreshed (such as updating and expanding event elements, etc.); otherwise, a new target event is created. Obviously, the target event has a life cycle (from the creation of the target event until the state of the target event no longer changes).

[0128] Then, in the process of tracking the state of the target event, an event warning can be performed. For example, the alarm score of the target event is determined based on the statistical information of the target event and the importance of the target event. The importance indicates that the more important the target event is, the larger the alarm score is. The statistical information indicates that the more statistics of the target event are, the larger the alarm score is. If the alarm score is greater than the threshold, an event warning for the target event is triggered.

[0129] Exemplarily, for each target event, during the state tracking process of the target event, statistical information of the target event can be detected, and the statistical information includes the statistics of each statistical period (i.e., the count of the statistical period). For example, if target event A occurs 10 times within statistical period 1, the statistics of statistical period 1 is 10; if target event A occurs 8 times within statistical period 2, the statistics of statistical period 2 is 8.

[0130] In addition, the statistical information may also include the year-on-year number of the statistical period. For example, when the statistical number of statistical period 1 is 10 and the statistical number of statistical period 2 is 8, the year-on-year number of statistical period 2 is -2, which means that the statistical number is 2 less than that of the previous statistical period. Of course, the above is only an example of statistical information and is not limited to this.

[0131] In addition, the statistical information of the target event can be normalized to obtain the statistical score value of the target event. For example, the larger the statistical amount of the statistical period, the larger the statistical score value of the target event in the statistical period. The larger the year-on-year amount of the statistical period, the larger the statistical score value of the target event in the statistical period.

[0132] Exemplarily, for each target event, the importance of the target event can also be determined. For example, a mapping relationship between event type and importance can be preconfigured. The importance of the target event can be obtained by querying the mapping relationship based on the event type of the target event, and the importance score of the target event can be determined based on the importance. For another example, the importance score of the target event can be given by a logistic regression (LogisticRegression) binary classifier, and there is no limitation on this implementation process.

[0133] Exemplarily, based on the statistical score value and the importance score value, the alarm score of the target event in a certain statistical period can be determined. If the alarm score is greater than a threshold, an event warning for the target event is triggered.

[0134] After triggering the event warning for the target event, the warning push can be performed, that is, the warning information of the target event is sent to the target device. For example, if the target event is target event A, the warning information of target event A (such as the event type is counterfeit complaint, the category is shoes, the store is xx store, the product is A sports shoes, the event time is time a1, time a2, etc., and 10 warnings occurred within statistical cycle 1, etc., and there is no restriction on the content of this warning information) can be sent to the target device.

[0135] For example, in an e-commerce scenario, the alarm information of the target event A can be sent to the target device via SMS, email, work notification, customer service robot, etc., and there is no restriction on this push method.

[0136] Based on the above technical solution, in the embodiment of the present application, the target neural network can be used to identify and extract event types and event elements, which effectively improves the utilization rate of context, helps to solve the problem of ambiguity, and upgrades the warning from the word and phrase granularity to the warning based on the "event" level, greatly enriching the information dimension of the warning result, improving readability, and being able to present information to users in a relatively complete and comprehensive manner. Moreover, there is no need for analysts to conduct in-depth analysis of the original text data of the e-commerce platform, solving the problems of heavy workload of analysts and easy analysis errors. A fully structured information extraction method can be used to represent the amount of information of the event itself. By extracting event types and event elements, the information can be presented to users in a relatively complete and comprehensive manner. These structured information include elements such as the subject, the object, the relationship between the subject and the object, and the state on the timeline. The event itself can be tracked in a full life cycle manner, and the different states of the event, such as the start, development, and extinction, and the switching between states can be dynamically updated. The event can be normalized and expressed by methods such as similarity calculation normalization and element entity unification, which can effectively avoid the problem of too fine granularity of the key phrase method.

[0137] Based on the same application concept as the above method, another event processing method is also proposed in the embodiment of the present application, see Figure 4 FIG. 4 is a flow chart of the method, which may include:

[0138] Step 401, obtaining the event type corresponding to the original text data and the event element corresponding to the event type.

[0139] Exemplarily, at least one trigger word can be obtained based on the original text data, and the at least one trigger word can be processed based on the trained target neural network to obtain the event type corresponding to the original text data and the event element corresponding to the event type. Of course, the implementation method of the target neural network is only an example and is not limited to this, as long as the event type and event element can be obtained.

[0140] Step 402: determine the target event according to the event type and the event element.

[0141] Exemplarily, the normalization element corresponding to the event element may be determined based on the type of the event element; then, the target event may be determined according to the event type and the normalization element.

[0142] Step 403: Process the target event according to the statistical information of the target event.

[0143] Exemplarily, the alarm score of the target event can be determined based on the statistical information of the target event; wherein, the more statistics of the target event represented by the statistical information, the larger the alarm score; if the alarm score is greater than a threshold, the alarm information of the target event can be sent to the target device.

[0144] Exemplarily, regarding step 402 and step 403, reference may be made to the above embodiments, which will not be described in detail here.

[0145] In one example, the above execution order is just an example given for the convenience of description. In practical applications, the execution order between the steps can also be changed, and there is no limitation on this execution order. Moreover, in other embodiments, the steps of the corresponding method are not necessarily executed in the order shown and described in this specification, and the steps included in the method may be more or less than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; multiple steps described in this specification may also be combined into a single step for description in other embodiments.

[0146] It should be further explained that the event handling method described in this specification can be applied not only in the e-commerce field, but also in various industries such as telecommunication services, government affairs, finance, education, entertainment, health, and tourism.

[0147] Based on the same application concept as the above method, an event processing device is also provided in the embodiment of the present application, such as Figure 5 FIG. 1 is a structural diagram of an event processing device, wherein the device comprises:

[0148] An acquisition module 51, configured to acquire at least one trigger word according to the original text data;

[0149] A processing module 52, configured to process the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0150] The determination module 53 is used to determine the target event according to the event type and the event element; the processing module 52 is also used to process the target event according to the statistical information of the target event.

[0151] The processing module 52 processes the at least one trigger word based on the trained target neural network to obtain the event type corresponding to the original text data and the event element corresponding to the event type, and is specifically used to: input at least one trigger word into the target neural network, so that the target neural network determines the event type trigger word based on the at least one trigger word, determines the event type corresponding to the original text data based on the event type trigger word, determines the event element trigger word based on the at least one trigger word, and determines the event element corresponding to the event type based on the event element trigger word.

[0152] When determining the target event according to the event type and the event element, the determination module 53 is specifically used to: determine the normalization element corresponding to the event element based on the type of the event element;

[0153] A target event is determined according to the event type and the normalization element.

[0154] The determination module 53 is specifically used to determine the normalization element corresponding to the event element based on the type of the event element: if the type of the event element is a product domain type, then determine the normalization element corresponding to the event element based on the product knowledge graph; wherein the product knowledge graph is pre-configured, and the product knowledge graph includes a mapping relationship between the event element and the normalization element.

[0155] The determination module 53 determines the normalized element corresponding to the event element based on the type of the event element, and is specifically used to: if the type of the event element is a general domain type, determine the target element entity corresponding to the event element from the element entity based on the similarity between the event element and the element entity;

[0156] Based on the generated element dictionary, a normalized element corresponding to the target element entity is determined; wherein the element dictionary includes a mapping relationship between a normalized element and at least one element entity.

[0157] When the processing module 52 processes the target event according to the statistical information of the target event, it is specifically used to: determine the alarm score of the target event according to the statistical information of the target event; wherein, when the statistical information indicates that the statistical amount of the target event is more, the alarm score is greater;

[0158] If the alarm score is greater than a threshold, the alarm information of the target event is sent to the target device.

[0159] When the processing module 52 determines the alarm score of the target event according to the statistical information of the target event, it is specifically used to: determine the alarm score of the target event according to the statistical information of the target event and the importance of the target event; the importance indicates that the more important the target event is, the larger the alarm score is.

[0160] Based on the same application concept as the above method, the embodiment of the present application further provides an event processing device, including: a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a plurality of computer instructions, and the processor performs the following processing when executing the computer instructions:

[0161] Acquire at least one trigger word according to the original text data;

[0162] Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0163] Determine a target event according to the event type and the event element;

[0164] The target event is processed according to the statistical information of the target event.

[0165] The present application also provides a machine-readable storage medium, on which a number of computer instructions are stored; when the computer instructions are executed, the following processing is performed:

[0166] Acquire at least one trigger word according to the original text data;

[0167] Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type;

[0168] Determine a target event according to the event type and the event element;

[0169] The target event is processed according to the statistical information of the target event.

[0170] See also Figure 6As shown, it is a structural diagram of the event processing device proposed in the embodiment of the present application. The event processing device 60 may include: a processor 61, a network interface 62, a bus 63, and a memory 64. The memory 64 may be any electronic, magnetic, optical or other physical storage device, which may contain or store information, such as executable instructions, data, etc. For example, the memory 64 may be: RAM (Radom Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid state hard disk, any type of storage disk (such as optical disk, DVD, etc.).

[0171] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email transceiver, a game console, a tablet computer, a wearable device or a combination of any of these devices.

[0172] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0173] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0174] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0175] Moreover, these computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0176] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0177] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. An event processing method, characterized in that: The method comprises: Acquire at least one trigger word according to the original text data; Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type; Determine a target event according to the event type and the event element; Processing the target event according to the statistical information of the target event; Wherein, determining the target event according to the event type and the event element includes: Based on the type of the event element, determine the normalized element corresponding to the event element; if the type of the event element is a commodity domain type, determine the normalized element corresponding to the event element based on the commodity knowledge graph; or, if the type of the event element is a general domain type, determine the target element entity corresponding to the event element from the element entities based on the similarity between the event element and the element entity; determine the normalized element corresponding to the target element entity based on the generated element dictionary; A target event is determined according to the event type and the normalization element.

2. The method according to claim 1, characterized in that: The processing of the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type includes: The at least one trigger word is input into the target neural network so that the target neural network determines an event type trigger word based on the at least one trigger word, determines the event type corresponding to the original text data based on the event type trigger word, determines an event element trigger word based on the at least one trigger word, and determines the event element corresponding to the event type based on the event element trigger word.

3. The method according to claim 2, characterized in that The determining of the event element trigger word based on the at least one trigger word, and determining the event element corresponding to the event type based on the event element trigger word, includes: An element role corresponding to the event type is determined based on the target neural network, an event element trigger word corresponding to the element role is determined based on the at least one trigger word, and an event element corresponding to the element role is determined based on the event element trigger word.

4. The method according to any one of claims 1 to 3, characterized in that: Before the at least one trigger word is processed based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type, the method further includes: Acquire training text data, and acquire at least one trigger word according to the training text data; The initial neural network is trained based on at least one trigger word corresponding to the training text data to obtain a trained target neural network; wherein the target neural network includes a mapping relationship between event type trigger words and event types, and a mapping relationship between event element trigger words and event elements; The event type trigger word is a trigger word used to determine the event type among the at least one trigger word; The event element trigger word is a trigger word used to determine an event element among the at least one trigger word.

5. The method according to claim 1, characterized in that The product knowledge graph is pre-configured, and the product knowledge graph includes a mapping relationship between event elements and normalized elements.

6. The method according to claim 1, characterized in that The element dictionary includes a mapping relationship between a normalized element and at least one element entity.

7. The method according to claim 6, characterized in that Before determining the normalized element corresponding to the target element entity based on the generated element dictionary, the method further includes: Acquire multiple element entities, cluster the multiple element entities based on a density clustering algorithm to obtain an element set, wherein the element set includes at least one element entity; generate the element dictionary based on a normalized element corresponding to the element set and at least one element entity included in the element set.

8. The method according to claim 1, characterized in that The processing of the target event according to the statistical information of the target event includes: Determining an alarm score of the target event according to statistical information of the target event; wherein the more statistics of the target event represented by the statistical information, the greater the alarm score; If the alarm score is greater than a threshold, the alarm information of the target event is sent to the target device.

9. The method according to claim 8, characterized in that The determining the alarm score of the target event according to the statistical information of the target event includes: The alarm score of the target event is determined according to the statistical information of the target event and the importance of the target event; wherein the importance indicates that the more important the target event is, the greater the alarm score is.

10. An event processing method, characterized in that: The method comprises: Acquire an event type corresponding to the original text data and an event element corresponding to the event type; Determine a target event according to the event type and the event element; Processing the target event according to the statistical information of the target event; Wherein, determining the target event according to the event type and the event element includes: Based on the type of the event element, determine the normalized element corresponding to the event element; if the type of the event element is a commodity domain type, determine the normalized element corresponding to the event element based on the commodity knowledge graph; or, if the type of the event element is a general domain type, determine the target element entity corresponding to the event element from the element entities based on the similarity between the event element and the element entity; determine the normalized element corresponding to the target element entity based on the generated element dictionary; A target event is determined according to the event type and the normalization element.

11. The method according to claim 10, characterized in that The processing of the target event according to the statistical information of the target event includes: Determining an alarm score of the target event according to statistical information of the target event; wherein the more statistics of the target event represented by the statistical information, the greater the alarm score; If the alarm score is greater than a threshold, the alarm information of the target event is sent to the target device.

12. An event processing device, characterized in that: The device comprises: An acquisition module, used for acquiring at least one trigger word according to the original text data; A processing module, configured to process the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type; A determination module, used for determining a target event according to the event type and the event element; The processing module is further used to process the target event according to the statistical information of the target event; Among them, when the determination module determines the target event according to the event type and the event element, it is specifically used to: determine the normalized element corresponding to the event element based on the type of the event element; if the type of the event element is a commodity domain type, then determine the normalized element corresponding to the event element based on the commodity knowledge graph; or, if the type of the event element is a general domain type, then determine the target element entity corresponding to the event element from the element entity based on the similarity between the event element and the element entity; determine the normalized element corresponding to the target element entity based on the generated element dictionary; determine the target event according to the event type and the normalized element.

13. An event processing device, characterized in that: include: A processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a plurality of computer instructions, and when the processor executes the computer instructions, the processor performs the following processing: Acquire at least one trigger word according to the original text data; Processing the at least one trigger word based on the trained target neural network to obtain an event type corresponding to the original text data and an event element corresponding to the event type; Determine a target event according to the event type and the event element; Processing the target event according to the statistical information of the target event; Wherein, determining the target event according to the event type and the event element includes: Based on the type of the event element, determine the normalized element corresponding to the event element; if the type of the event element is a commodity domain type, determine the normalized element corresponding to the event element based on the commodity knowledge graph; or, if the type of the event element is a general domain type, determine the target element entity corresponding to the event element from the element entities based on the similarity between the event element and the element entity; determine the normalized element corresponding to the target element entity based on the generated element dictionary; A target event is determined according to the event type and the normalization element.

Images