CN117454003A - Sample data generation method, model training method and search result determination method - Google Patents

Abstract

The disclosure provides a sample data generation method, relates to the technical field of artificial intelligence, and particularly relates to the technical field of deep learning and intelligent recommendation. The specific implementation scheme is as follows: identifying a core word of the target object from the search statement; identifying industry information of the target object according to the core word; determining the search requirement of a search sentence according to at least one of the core word and the industry information, and taking the search requirement as a requirement label of the search sentence, wherein the requirement label comprises one of a source searching class, a content class and other classes; and determining the search statement with the demand label as sample data. The disclosure also provides a training method of the deep learning model, a search result determining method, a device, electronic equipment and a storage medium.

Description

Sample data generation method, model training method and search result determination method

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to the technical field of deep learning and intelligent recommendation. More specifically, the present disclosure provides a sample data generating method, a training method of a deep learning model, a search result determining method, an apparatus, an electronic device, and a storage medium.

Background

When a user inputs a search term (Query) on a search platform, the user's search needs need to be identified in order to recommend more accurate page information to the user.

Disclosure of Invention

The present disclosure provides a sample data generating method, apparatus, device, and storage medium.

According to a first aspect, there is provided a sample data generation method, the method comprising: identifying a core word of the target object from the search statement; identifying industry information of the target object according to the core word; determining the search requirement of a search sentence according to at least one of the core word and the industry information, and taking the search requirement as a requirement label of the search sentence, wherein the requirement label comprises one of a source searching class, a content class and other classes; determining search statement with demand label as sample data

According to a second aspect, there is provided a training method of a deep learning model, the method comprising: acquiring sample data, wherein the sample data is generated according to the sample generation method; and training the deep learning model by using the sample data to obtain the deep learning model for identifying the search requirement.

According to a third aspect, there is provided a search result determination method, the method comprising: acquiring an input sentence; identifying a search requirement for the input sentence using the trained deep learning model, wherein the search requirement includes one of a sourcing class, a content class, and other classes; generating a search result page according to the search requirement; the deep learning model is obtained by training according to the training method of the deep learning model.

According to a fourth aspect, there is provided a sample data generating apparatus comprising: the core word determining module is used for identifying the core word of the target object from the search statement; the industry information determining module is used for identifying the industry information of the target object according to the core word; the search requirement determining module is used for determining the search requirement of the search sentence according to at least one of the core word and the industry information and taking the search requirement as a requirement label of the search sentence, wherein the requirement label comprises one of a source searching class, a content class and other classes; and a sample data generation module for determining the search statement with the requirement label as sample data.

According to a fifth aspect, there is provided a training apparatus of a deep learning model, the apparatus comprising: the sample acquisition module is used for acquiring sample data, wherein the sample data is generated according to the sample data generation device; and the training module is used for training the deep learning model by using the sample data to obtain the deep learning model for identifying the search requirement.

According to a sixth aspect, there is provided a search result determination apparatus comprising: the input sentence determining module is used for acquiring an input sentence; a processing module for identifying a search requirement for the input sentence using the trained deep learning model, wherein the search requirement includes one of a sourcing class, a content class, and other classes; the search result determining module is used for generating a search result page according to the search requirement; the deep learning model is obtained by training according to the training device of the deep learning model.

According to a seventh aspect, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.

According to an eighth aspect, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a method provided according to the present disclosure.

According to a ninth aspect, there is provided a computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which, when executed by a processor, implements a method provided according to the present disclosure.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram of an exemplary system architecture to which at least one of a sample data generation method, a training method of a deep learning model, a search result determination method may be applied, according to one embodiment of the present disclosure;

FIG. 2 is a flow chart of a sample data generation method according to one embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a search requirement identification method according to one embodiment of the present disclosure;

FIG. 4 is a flow chart of a training method of a deep learning model according to one embodiment of the present disclosure;

FIG. 5 is a flow chart of a search result determination method according to one embodiment of the present disclosure;

FIG. 6 is a schematic diagram of a method of determining search requirements according to one embodiment of the present disclosure;

FIG. 7 is a block diagram of a sample data generating device according to one embodiment of the present disclosure;

FIG. 8 is a block diagram of a training apparatus of a deep learning model according to one embodiment of the present disclosure;

FIG. 9 is a block diagram of a search result determination apparatus according to one embodiment of the present disclosure;

fig. 10 is a block diagram of an electronic device of at least one of a sample data generating method, a training method of a deep learning model, a search result determining method according to one embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

Under different application scenes, the division modes of the demand categories are different. Under the search scenario of the e-commerce platform, there are two main categories of interest: a source-seeking class and a content class. The source searching type refers to the purchasing price inquiring requirement of the user, the content type refers to the information searching requirement of the user, and the searching requirements except the two requirements are classified into other types.

Currently, search demand identification methods include the following several implementations.

One method of search requirement identification is to train a classification model using manually labeled data. For example, by manually labeling the demand categories of the sample Query, classification models for identifying the demand categories are trained using the labeling data. The manual labeling method is high in cost, a large number of samples are difficult to obtain, and therefore the accuracy and recall rate of the trained model are difficult to improve.

One method of search requirement identification is to directly infer search requirements using a pre-trained language model. For example, the pre-trained intention classification model is used for identifying the demands of purchase, price inquiry, question and answer, text, video and the like of the Query, and then post-processing is carried out to obtain the intention classification of the Query. However, the function of the pre-training model often does not completely match with the actual classification requirement, and it is difficult to directly and accurately identify the requirement class.

ToB (Business to Business, enterprise-oriented) business refers to business targeting enterprise-level users, such as wholesale channel purchasing business, and the like. The ToC (To Customer) service refers To a service targeting individual users.

The identification of the demand category in the ToB/ToC service scene has important significance, and after the demand category is identified, the page can be displayed and produced for the user in a targeted manner, so that the user experience, the coverage rate of the platform, the flow ratio and the diversion magnitude are improved.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the user accord with the regulations of related laws and regulations, and the public order colloquial is not violated.

In the technical scheme of the disclosure, the authorization or consent of the user is obtained before the personal information of the user is obtained or acquired.

FIG. 1 is a schematic diagram of an exemplary system architecture to which at least one of a sample data generation method, a training method of a deep learning model, a search result determination method may be applied, according to one embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.

As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. The terminal devices 101, 102, 103 may be a variety of electronic devices including, but not limited to, smartphones, tablets, laptop portable computers, and the like.

At least one of the sample data generating method and the search result determining method provided by the embodiments of the present disclosure may be generally performed by the terminal devices 101, 102, 103. Accordingly, at least one of the sample data generating means and the search result determining means provided by the embodiments of the present disclosure may be generally provided in the terminal device 101, 102, 103.

The training method of the deep learning model provided by the embodiments of the present disclosure may be generally performed by the server 105. Accordingly, the training apparatus of the deep learning model provided by the embodiments of the present disclosure may be generally disposed in the server 105.

Fig. 2 is a flow chart of a sample data generation method according to one embodiment of the present disclosure.

As shown in fig. 2, the sample data generating method 200 includes operations S210 to S240.

In operation S210, a core word of a target object is identified from a search sentence.

The search statement, query, may be a collection of search records for a large number of users. The core words in Query include entity words of the target object (e.g., the article body), entity words associated with the target object, modifier words, etc., such as brand names, place names, collar names, brand recommendations, etc. The target object may be an item (e.g., a mechanical product, an electronic product, etc.) or a service (e.g., an advertisement, a marketing service, etc.). Specifically, a core word in the Query can be identified by using a word labeling model.

In operation S220, industry information to which the target object belongs is identified according to the core word.

The industry to which the target object belongs can be identified based on the core word using a pre-trained industry classification model. For example, core words of the target object (e.g., target object body, brand name, domain name, etc.) may be input into a pre-trained industry classification model to obtain the industry to which the target object belongs.

Specifically, the primary industry, the secondary industry, and the like of the target object can also be finely identified. For example, the primary industry may be manufacturing, the secondary industry may be machine manufacturing, and so on.

In operation S230, a search requirement of a search sentence is determined as a requirement tag of the search sentence according to at least one of the core word and industry information.

According to the embodiment of the disclosure, intent recognition can be performed on the search statement according to the core word, so that the search requirement of the search statement is obtained; and determining the search requirement of the search statement according to the matching rule corresponding to the industry information.

For example, a pre-trained intent recognition model may be used to perform intent recognition based on core words, recognize the user's needs for purchase, price-ask, question-answer, text, video, etc., determine the purchase and price-ask needs as a sourcing class, and determine the question-answer, text, and video needs as a content class.

For example, a corresponding content recognition rule and a source-seeking recognition rule may be set for each industry, and for a Query whose target object belongs to a certain industry, whether the Query is a content class or a source-seeking class is determined using the content recognition rule and the source-seeking recognition rule.

For example, for the mechanical device industry, content recognition rules may correspond to a content scope that may include basic attributes such as brand recommendations, feature functions, model sizes, definition principles, and the like. If a Query or a core word of a Query falls within these basic attribute ranges, it may be determined that the Query belongs to the content class. The content scope may also include Query attributes of a Query, i.e., whether the Query expresses a Query, and if so, it may be determined that the Query belongs to the content class.

For example, for the mechanical device industry, the source identification rule may correspond to a source range, which may include product customization recycling, after-market maintenance information, and so forth. If a Query or a core word of a Query falls within a sourcing range, it may be determined that the Query belongs to a sourcing class.

For example, for Query that does not fall within the content range, nor the sourcing range, the demand type may be determined to be of other classes.

In operation S240, a search statement having a requirement tag is determined as sample data.

After obtaining the demand label of the Query (sourcing class, content class, other class), the Query with the demand label may be taken as a sample. And training an end-to-end classification model by using the sample, wherein the classification model has the capability of identifying the demand category of the Query.

According to the embodiment of the disclosure, the core words in the search sentences and the corresponding industry information are identified, the search requirements are determined based on the core words and the industry information and used as the requirement labels, the sample data are obtained, the sample data are convenient to use for training of the model, and the deep learning model for identifying the requirement category is obtained, so that the identification efficiency, recall rate and accuracy of the requirement category are improved.

Fig. 3 is a schematic diagram of a search requirement identification method according to one embodiment of the present disclosure.

As shown in fig. 3, the present embodiment includes a Query requirement recognition module 301 based on core words, a Query requirement recognition module 302 based on industry information, a special class Query recognition module 303, and a source-seeking class Query recognition module 304.

The respective identification modules are explained below.

First, the Query requirement recognition module 301 based on the core word will be described. Core words in Query can be identified using the lexical labeling model. The core word is input into a pre-trained intention recognition model, requirements of purchase, price inquiry, question and answer, text, video and the like can be obtained, the requirements of purchase and price inquiry are divided into source searching types, and the question and answer, text and video are divided into content types.

The Query requirement identification module 302 based on industry information is described below.

According to an embodiment of the present disclosure, the industry information includes a first type of industry, a second type of industry, and a third type of industry, the first type of industry including an industry strongly associated with enterprise-oriented business, the third type of industry including an industry strongly associated with personal-oriented business, the second type of industry including an industry cross-associated with enterprise-oriented business and personal-oriented business; determining search requirements of the search statement according to the matching rules corresponding to the industry information comprises: determining the search requirement of a search statement as one of a source searching class and a content class according to matching rules corresponding to various industries; and determining the search requirement of the search statement as other classes according to the object range irrelevant to all types of industries.

The first type of industry includes machinery, agriculture, forestry, animal husbandry, etc. obviously belonging to the ToB type of industry. The broadest range of content recognition rules may be set for the first type of industry. For example, the first content range corresponding to the first type industry includes basic attribute information (brand recommendation, feature function, model size, definition principle) of the target object, quotation information, trouble shooting information, usage information (used methods and rules), and query attributes of the search sentence. For example, regular matching may be used to determine whether a Query of a first industry type belongs to a first content range, and if so, to divide into content classes. For another example, a pre-trained intent recognition model may be used to identify whether Query is questioned, and if so, to classify into content classes. In one example, for all Query expressing questions, the content class may be divided.

The second type of industry includes the 3C, furniture, second hand truck, and other industries. The second type industry may be provided with identification rules for a second content range. For example, the second content range includes basic attribute information (brand recommendation, feature function, model size, definition principle), quotation information, trouble shooting information, and usage information (used methods and rules) of the target object in the Query. Regular matching may be used to determine whether a Query of a second industry type belongs to the first content range and if so, to divide into content classes.

The third type of industry type includes industries biased toward the ToC type, such as food and beverage, consumer goods, and the like. A third type of industry may be provided with a strictest third content scope that includes basic attribute information (brand recommendation, feature function, model size, definition principles) of a target object in Query. Regular matching may be used to determine whether a Query of a second industry type belongs to the first content range and if so, to divide into content classes.

The Query of the content class in the industries of the above types is related to the goods or services in the industry. For Query independent of goods or services, other classes may be divided. For example, object scopes unrelated to various types of industries include entertainment, consultation, movies, and the like. If the core word of the Query falls within these ranges, the class of requirements of the Query can be determined to be other classes.

The embodiment can identify a secondary industry to which the target object in the Query belongs by using the industry classification model. One of the purposes of obtaining industry classifications herein is that because Query of source-seeking class and content class should be related to an article or service, query under industry unrelated to an article or service can be culled (i.e., categorized as other class) by utilizing industry classifications for Query unrelated to an article or service. Another object is that the standards of Query in different industries are different, for example, the usage method of the mechanical device belongs to the content class, and the usage method of the daily consumer product does not belong to the content class, so that after the industry classification result is obtained, a refined content class identification rule can be set for different industries.

Next, the special class Query recognition module 303 will be described.

According to the embodiment of the disclosure, each type of industry may be provided with a corresponding special category range, and the search requirement of the search sentence of which the target object belongs to the special category range is determined as other categories according to the special category range corresponding to each type of industry.

In embodiments that identify Query demand category modules based on industry information, some of the Query may be identified as being related to items/services in a particular industry, but in fact, because the category is specific and will not be sold by the platform (e.g., a special function vehicle such as an ambulance, etc.), it is desirable to identify and categorize these Query into other categories. Query under a particular industry may be filtered using intent recognition models or regular matching methods already existing in that industry. For example, using the existing ToB Query recognition model in the first type industry and the regular matching method, the Query is recognized, and if the recognition is successful, the target objects of the Query belong to the normal goods or services in the ToB industry. If the identification is not successful, the target object of the Query belongs to a special class in the ToB industry, and the demand class of the Query is determined to be other classes.

Next, the source-seeking class Query recognition module 304 will be described.

According to the embodiment of the disclosure, whether the search requirement of the search statement is a source searching class is determined according to the source searching matching rule corresponding to various types of industries; and determining a search requirement of the search sentence as a source searching class in response to the search sentence being one core word or a combination of a plurality of core words, wherein the core word is related to one of an article entity and a service entity in various types of industries.

In the embodiment of the Query demand recognition module based on the core word, although an existing pre-trained intention recognition model is used, the source-seeking demands such as purchase, price polling and the like contained in the Query can be recognized, but the accuracy of the model is problematic. Therefore, in order to improve the accuracy of identifying the source-seeking class Query, the embodiment synthesizes a rule matching method to identify the source-seeking class Query so as to improve the accuracy and recall rate of the source-seeking class Query.

For example, a higher threshold may be used when using a pre-trained intent model to identify purchasing, price-enquiring needs. Thus, partial recall rate of the source searching class is sacrificed first so as to ensure the accuracy of identification. And then, based on the source searching matching rule, carrying out recall of the source searching class Query by using regular matching, and improving the recall rate. The range to which the source-seeking matching rule corresponds may include product customization recycling, repair after-market related information. These are difficult to identify by pre-trained intent models, and these Query can be identified using regular matching. Furthermore, query, which has only a core part and a core word part related to an article or service, can be classified as a sourcing class. When a Query has only a core word portion and the Query is a noun phrase related to goods/services, the Query belongs to a source-seeking class with a high probability, so that the Query can be classified as the source-seeking class.

It should be noted that, for Query with inconsistent demand categories obtained by using different recognition methods or matching methods. For example, for the Query containing maintenance related information in the ToB industry belongs to the content matching range and also belongs to the source searching matching range, arbitration may be performed based on a pre-training intention recognition model, for example, the confidence that the Query output by the intention recognition model belongs to the source searching class and the content class is determined, and a class with high confidence is selected as a requirement label of the Query.

According to the method, the requirements of the sample Query are classified by using a plurality of pre-training models and a regular matching method, and the labels of the sample Query are obtained, so that the problem of poor recognition effect of a single pre-training model is avoided, and the recognition accuracy of the Query of each requirement type is improved.

Fig. 4 is a flowchart of a training method of a deep learning model according to one embodiment of the present disclosure.

As shown in fig. 4, the training method 400 of the deep learning model includes operations S410 to S420.

In operation S410, sample data is acquired.

In operation S420, a deep learning model is trained using the sample data, resulting in a deep learning model for identifying search requirements.

Sample data is generated according to the sample data generation method described above. The sample data has a demand label, the demand label for each sample data being one of a sourcing class, a content class, and other classes.

Training is performed on the basis of the pre-training language model by using sample data, so that a trained deep learning model is obtained, and the trained deep learning model has the capability of identifying search requirements. The pre-trained language model may be a Bert model,

one of the benefits of model training using Query with demand category labels is that classification speed can be increased. The sample generation process can give high-quality classification results for Query, but multiple pre-training models and regular matching methods are required to be used in series during prediction, so that the recognition efficiency is low. The embodiment uses the generated sample to train a new model of end-to-end prediction, and the model has high inference speed and can be used for online use. Another beneficial effect is to promote recall rate and accuracy rate of source-seeking class and content class Query. The classification model can be obtained by training on the basis of a pre-trained language model, and the language model encodes the semantic similarity of the text in the pre-training process, so that after fine tuning, the classification rule of a sample can be learned to a great extent, and the semantic smoothness can be added for the classification standard. For sourcing classes and content class Query that are missing during the sample generation phase, some can be correctly identified after model training. Therefore, training the new model can promote the recognition recall rate and accuracy rate for the source-seeking class and the content class.

Fig. 5 is a flow chart of a search result determination method according to one embodiment of the present disclosure.

As shown in fig. 5, the training method 500 of the deep learning model includes operations S510 to S520.

In operation S510, an input sentence is acquired.

In operation S520, a search requirement of the input sentence is identified using the trained deep learning model. The deep learning model is obtained by training according to the training method of the deep learning model.

In operation S530, a search result page is generated according to the search requirement.

The input statement Query may be entered by a user at the search platform. The input sentence is input into the trained deep learning model, and a search requirement of the input sentence can be obtained, wherein the search requirement comprises one of a source searching class, a content searching class and other classes.

According to the search requirement of the input statement Query, pages can be displayed and produced for users in a targeted manner. For example, related pages for purchase, price inquiry, brand recommendation, etc. may be produced for the input statement Query of the sourcing class. Pages of the content introduction class may be produced for the input statement Query of the content class.

According to the embodiment, the search requirement is determined according to the input sentences, and pages can be displayed and produced for the user in a targeted manner, so that user experience, platform coverage rate, flow rate duty ratio and diversion magnitude are improved.

According to an embodiment of the present disclosure, the method further comprises: determining an input object in an input sentence; determining the business type of the input statement according to the industry information of the input object, wherein the business type comprises an enterprise-oriented business type, a personal-oriented business type and a cross business type; and generating a search result page according to the service type and the search requirement of the input sentence.

For example, the pre-trained industry classification model may be used to identify industry information to which the Query belongs, and then determine the business type of the Query according to the industry type. The industry types may include a first type of industry including an industry strongly associated with an enterprise-oriented business (ToB industry), a second type of industry including an industry strongly associated with a personal-oriented business (ToC industry), and a third type of industry including an industry cross-associated with an enterprise-oriented business and a personal-oriented business.

If the Query belongs to the industry as the first industry type, determining that the Query service type is the ToB service type. And if the Query belongs to the industry which is the second industry type, determining that the Query service type is the cross service type. If the Query belongs to the industry of the third industry type, determining that the Query service type is the ToC service type.

After identifying the business type of Query, a more refined and accurate page can be produced for the user. For example, for a Query that is of ToB business type and is a sourcing requirement, more pages containing wholesale purchase channel information may be produced. For Query, which is of the ToC business type and is a sourcing requirement, more pages containing non-wholesale channel information can be produced.

According to the implementation, according to the input sentences and the service types and the demand categories, pages are displayed and produced for the user in a targeted manner, the user searching demands can be met more precisely, and therefore user experience is improved.

FIG. 6 is a schematic diagram of a method of determining search requirements according to one embodiment of the present disclosure.

As shown in fig. 6, the present embodiment includes a sample generation stage, a model fine tuning stage, and a demand category prediction stage. And obtaining a Query demand label by using the Query demand classification method, and generating a sample Query. The model after fine tuning is obtained by using a sample Query fine tuning pre-trained language model (for example, a Bert model). And carrying out demand recognition on the Query newly input by the user by using the fine-tuning model to obtain the search demand category of the user.

Fig. 7 is a block diagram of a sample data generating device according to one embodiment of the present disclosure.

As shown in fig. 7, the sample data generating apparatus 700 includes a core word determining module 701, an industry information determining module 702, a search demand determining module 703, and a sample data generating module 704.

The core word determining module 701 is configured to identify a core word of a target object from the search statement.

The industry information determining module 702 is configured to identify, according to the core word, industry information to which the target object belongs.

The search requirement determining module 703 is configured to determine, according to at least one of the core word and the industry information, a search requirement of the search sentence as a requirement tag of the search sentence, where the requirement tag includes one of a source-seeking class, a content class, and other classes.

The sample data generation module 704 is configured to determine a search statement with a requirement tag as sample data.

The search requirement determination module 703 includes an intent recognition sub-module and a matching sub-module.

The intention recognition submodule is used for carrying out intention recognition on the search sentences according to the core words to obtain the search requirements of the search sentences.

And the matching sub-module is used for determining the search requirement of the search statement according to the matching rule corresponding to the industry information.

According to an embodiment of the present disclosure, the industry information includes a first type of industry, a second type of industry, and a third type of industry, the first type of industry including an industry strongly associated with enterprise-oriented business, the third type of industry including an industry strongly associated with personal-oriented business, the second type of industry including an industry cross-associated with enterprise-oriented business and personal-oriented business. The matching submodule comprises a first determining unit and a second determining unit.

The first determining unit is used for determining that the search requirement of the search statement is one of a source searching class and a content class according to matching rules corresponding to various types of industries.

The second determining unit is used for determining that the search requirement of the search statement is other classes according to the object range irrelevant to all types of industries.

The first determination unit includes a first content matching sub-unit, a second content matching sub-unit, and a third content matching sub-unit.

The first content matching subunit is configured to determine, for a search sentence of which the target object belongs to a first type industry, whether a search requirement of the search sentence is a content class based on a content matching rule in a first range, where the first range includes basic attribute information, market information, fault maintenance information, usage information, and query attribute information of the search sentence of the target object.

The second content matching subunit is configured to determine, according to a content matching rule in a second range, whether a search requirement of the search statement is a content class, where the second range includes basic attribute information, market information, fault maintenance information, and usage information of the target object, for the search statement that the target object belongs to a second type industry.

The third content matching subunit is configured to determine, according to a content matching rule of a third range, whether a search requirement of the search statement is a content class, where the third range includes basic attribute information of the target object, for a search statement that the target object belongs to a third type industry.

The first determination unit further includes a first source-seeking matching subunit and a second source-seeking matching subunit.

The first source searching matching subunit is used for determining whether the searching requirement of the searching statement is a source searching class according to the source searching matching rules corresponding to various types of industries.

And the second source searching matching subunit is used for responding to the search statement as a core word or a combination of a plurality of core words, and the core word is related to one of an article entity and a service entity in various industries, so that the search requirement of the search statement is determined to be a source searching class.

The matching sub-module further comprises a special class determining unit, which is used for determining the search requirement of the search statement of which the target object belongs to the special class range as other classes according to the special class range corresponding to each type of industry.

Fig. 8 is a block diagram of a training apparatus of a deep learning model according to one embodiment of the present disclosure.

As shown in fig. 8, the training apparatus 800 of the deep learning model includes a sample acquisition module 801 and a training module 802.

A sample acquisition module 801, configured to acquire sample data. Wherein the sample data is generated by the sample data generating device.

Training module 802 for training a deep learning model using the sample data to obtain a deep learning model for identifying search needs.

Fig. 9 is a block diagram of a search result determination apparatus according to one embodiment of the present disclosure.

As shown in fig. 9, the search result determination apparatus 900 includes an input sentence acquisition module 901, a processing module 902, and a search result determination module 903.

The input sentence acquisition module 901 is configured to acquire an input sentence.

The processing module 902 is configured to identify a search requirement for the input sentence using the trained deep learning model, wherein the search requirement includes one of a sourcing class, a content class, and other classes. The trained deep learning model is obtained by using the deep learning model training device.

The search result determining module 903 is configured to generate a search result page according to a search requirement.

The deep learning model is obtained by training according to the training device of the deep learning model.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

Fig. 10 shows a schematic block diagram of an example electronic device 1000 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 10, the apparatus 1000 includes a computing unit 1001 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a Random Access Memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other by a bus 1004. An input/output (I/O) interface 1005 is also connected to bus 1004.

Various components in device 1000 are connected to I/O interface 1005, including: an input unit 1006 such as a keyboard, a mouse, and the like; an output unit 1007 such as various types of displays, speakers, and the like; a storage unit 1008 such as a magnetic disk, an optical disk, or the like; and communication unit 1009 such as a network card, modem, wireless communication transceiver, etc. Communication unit 1009 allows device 1000 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunications networks.

The computing unit 1001 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 1001 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The calculation unit 1001 performs the respective methods and processes described above, for example, at least one of a sample data generation method, a training method of a deep learning model, and a search result determination method. For example, in some embodiments, at least one of the sample data generation method, the training method of the deep learning model, the search result determination method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program may be loaded and/or installed onto device 1000 via ROM 1002 and/or communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of at least one of the sample data generating method, the training method of the deep learning model, and the search result determining method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform at least one of the sample data generation method, the training method of the deep learning model, the search result determination method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on 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.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed aspects are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims (23)

1. A sample data generation method, comprising:

identifying a core word of the target object from the search statement;

identifying industry information of the target object according to the core word;

determining a search requirement of the search statement as a requirement label of the search statement according to at least one of the core word and the industry information, wherein the requirement label comprises one of a source searching class, a content class and other classes; and

the search statement with the demand label is determined as sample data.

2. The method of claim 1, wherein the determining the search requirement of the search term according to at least one of the core word and the industry information, as a requirement tag of the search term, includes at least one of:

Performing intention recognition on the search sentences according to the core words to obtain search requirements of the search sentences;

and determining the search requirement of the search statement according to the matching rule corresponding to the industry information.

3. The method of claim 2, wherein the industry information comprises a first type of industry, a second type of industry, and a third type of industry, the first type of industry comprising an industry strongly associated with business oriented, the third type of industry comprising an industry strongly associated with personal oriented, the second type of industry comprising an industry cross-associated with business oriented and personal oriented; the determining the search requirement of the search statement according to the matching rule corresponding to the industry information comprises:

determining the search requirement of the search statement as one of a source searching class and a content class according to matching rules corresponding to various industries; and

and determining the search requirement of the search statement as other classes according to the object range irrelevant to all types of industries.

4. A method according to claim 3, wherein the matching rules comprise content matching rules; the determining that the search requirement of the search statement is one of a source searching class and a content class according to the matching rule corresponding to various types of industries comprises:

Determining whether the search requirement of a search statement is a content class or not based on a content matching rule of a first range aiming at a search statement of which a target object belongs to a first type industry, wherein the first range comprises basic attribute information, quotation information, fault maintenance information, use information and query attribute information of the search statement;

aiming at a search statement of which a target object belongs to a second type industry, determining whether the search requirement of the search statement is a content class or not based on a content matching rule of a second range, wherein the second range comprises basic attribute information, market information, fault maintenance information and use information of the target object;

and determining whether the search requirement of the search statement is a content class or not according to a content matching rule of a third range aiming at the search statement of which the target object belongs to a third type industry, wherein the third range comprises basic attribute information of the target object.

5. The method of claim 3 or 4, wherein the matching rules comprise sourcing matching rules; the determining that the search requirement of the search statement is one of a source searching class and a content class according to the matching rule corresponding to various types of industries comprises:

Determining whether the search requirement of the search statement is a source searching class according to a source searching matching rule corresponding to various industries; and

and determining the search requirement of the search sentence as a source searching class in response to the search sentence being one core word or a combination of a plurality of core words, wherein the core word is related to one of an article entity and a service entity in various industries.

6. The method of any one of claims 3 to 5, further comprising:

and determining the search requirement of the search statement of which the target object belongs to the special class range as other classes according to the special class range corresponding to various types of industries.

7. A training method of a deep learning model, comprising:

obtaining sample data, wherein the sample data is generated according to the method of any one of claims 1 to 6; and

and training a deep learning model by using the sample data to obtain the deep learning model for identifying the search requirement.

8. A search result determination method, comprising:

acquiring an input sentence;

identifying a search requirement for the input sentence using a trained deep learning model, wherein the search requirement includes one of a sourcing class, a content class, and other classes; and

Generating a search result page according to the search requirement;

wherein the deep learning model is trained according to the method of claim 7.

9. The method of claim 8, the method further comprising:

determining an input object in the input sentence; and

and determining the business type of the input statement according to the industry information of the input object, wherein the business type comprises an enterprise-oriented business type, a personal-oriented business type and a cross business type.

10. The method of claim 9, wherein the generating a search results page in accordance with the search requirement comprises:

and generating a search result page according to the service type and the search requirement of the input sentence.

11. A sample data generating apparatus comprising:

the core word determining module is used for identifying the core word of the target object from the search statement;

the industry information determining module is used for identifying the industry information of the target object according to the core word;

a search requirement determining module, configured to determine, according to at least one of the core word and the industry information, a search requirement of the search sentence as a requirement tag of the search sentence, where the requirement tag includes one of a source searching class, a content searching class, and other classes; and

And the sample data generation module is used for determining the search statement with the requirement label as sample data.

12. The apparatus of claim 11, wherein the search demand determination module comprises:

the intention recognition sub-module is used for carrying out intention recognition on the search statement according to the core word to obtain the search requirement of the search statement;

and the matching sub-module is used for determining the search requirement of the search statement according to the matching rule corresponding to the industry information.

13. The apparatus of claim 12, wherein the industry information comprises a first type of industry, a second type of industry, and a third type of industry, the first type of industry comprising an industry strongly associated with business oriented, the third type of industry comprising an industry strongly associated with personal oriented, the second type of industry comprising an industry cross-associated with business oriented and personal oriented; the matching submodule includes:

the first determining unit is used for determining that the search requirement of the search statement is one of a source searching class and a content class according to the matching rules corresponding to various industries; and

and the second determining unit is used for determining the search requirement of the search statement as other classes according to the object range irrelevant to all types of industries.

14. The apparatus of claim 13, wherein the matching rules comprise content matching rules; the first determination unit includes:

the first content matching subunit is used for determining whether the search requirement of the search statement is a content class or not according to a content matching rule of a first range aiming at a search statement of which a target object belongs to a first type industry, wherein the first range comprises basic attribute information, market information, fault maintenance information, use information and query attribute information of the search statement;

the second content matching subunit is used for determining whether the search requirement of the search statement is a content class or not according to a content matching rule of a second range aiming at the search statement of which the target object belongs to a second type industry, wherein the second range comprises basic attribute information, market information, fault maintenance information and use information of the target object;

and the third content matching subunit is used for determining whether the search requirement of the search statement is a content class or not according to a content matching rule of a third range aiming at the search statement of which the target object belongs to a third type industry, wherein the third range comprises basic attribute information of the target object.

15. The apparatus of claim 13 or 14, wherein the matching rules comprise sourcing matching rules; the first determination unit includes:

the first source searching matching subunit is used for determining whether the searching requirement of the searching statement is a source searching class according to the source searching matching rules corresponding to various industries; and

and the second source searching matching subunit is used for responding to the search statement as a core word or a combination of a plurality of core words, wherein the core word is related to one of an article entity and a service entity in various industries, and the search requirement of the search statement is determined to be a source searching class.

16. The apparatus of any of claims 13 to 15, wherein the matching sub-module further comprises:

and the special class determining unit is used for determining the search requirement of the search statement of which the target object belongs to the special class range as other classes according to the special class range corresponding to various types of industries.

17. A training device for a deep learning model, comprising:

a sample acquisition module for acquiring sample data, wherein the sample data is generated according to the apparatus of any one of claims 11 to 16; and

And the training module is used for training the deep learning model by using the sample data to obtain the deep learning model for identifying the search requirement.

18. A search result determination apparatus, comprising:

the input sentence determining module is used for acquiring an input sentence;

a processing module for identifying a search requirement for the input sentence using a trained deep learning model, wherein the search requirement includes one of a sourcing class, a content class, and other classes; and

the search result determining module is used for generating a search result page according to the search requirement;

wherein the deep learning model is trained from the apparatus of claim 17.

19. The apparatus of claim 18, the apparatus further comprising:

an input object determining module, configured to determine an input object in the input sentence; and

and the business type determining module is used for determining the business type of the input statement according to the business information of the input object, wherein the business type comprises an enterprise-oriented business type, a personal-oriented business type and a cross business type.

20. The apparatus of claim 19, wherein the search result determination module is configured to generate a search result page according to a service type and a search requirement of the input sentence.

21. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 10.

22. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 10.

23. A computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which, when executed by a processor, implements the method according to any one of claims 1 to 10.