CN113297471A

CN113297471A - Method and device for generating data object label and searching data object and electronic equipment

Info

Publication number: CN113297471A
Application number: CN202011255763.2A
Authority: CN
Inventors: 郭宗义; 王彬; 潘攀
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Singapore Holdings Pte Ltd
Priority date: 2020-11-11
Filing date: 2020-11-11
Publication date: 2021-08-24

Abstract

The embodiment of the application discloses a method, a device and electronic equipment for generating a data object label and searching a data object, wherein the method comprises the following steps: obtaining a classification model, wherein the classification model is obtained by training a training sample, the training sample comprises an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, and a tagging result of the image of the target data object is determined by analyzing a target user behavior record; determining a data object to be predicted and image information corresponding to the data object; and inputting the image information of the data object to be predicted into the classification model to obtain a corresponding text label. By the embodiment of the application, more effective training samples can be obtained, and then the labels are automatically generated for the data objects more effectively.

Description

Method and device for generating data object label and searching data object and electronic equipment

Technical Field

The present application relates to the field of data processing technologies, and in particular, to a method and an apparatus for generating a data object tag, a method and an apparatus for searching a data object, and an electronic device.

Background

In a search/recommendation scenario of an information service system for data objects (goods objects or services, etc.), various tags are usually marked on the data objects, such as property information of the goods objects, such as style, color, etc., sales information of goods, such as sales promotion, discount, etc., when a user searches the data objects through vocabularies, if a search word hits the tag information, the data objects with the tag information are preferentially returned due to higher association, so that the user can obtain information of the data objects closest to the original intention. In other words, the tag of the data object is a more fine-grained feature description of the data object, and has a key influence on improving the exposure, click and purchase rate of the data object.

The generation technology of the data object label can be generally divided into two categories of manual filling and algorithm automatic filling, and in the method based on the manual filling, a merchant user selects corresponding category information and adds corresponding text label description information when publishing the data object. However, the manual filling method is not accurate, and there are cases where addition is abandoned due to reasons such as uncertainty about how to add.

However, many methods for automatically filling the algorithm are performed, for example, based on the text features of the product, by constructing a text feature database, and by calculating the similarity between the text features (which can be usually extracted from the detail page) of the current data object and the text data features of the database, the tag is automatically generated, so as to be used in the specific scenes such as search and recommendation of the data object. However, in a cross-border scenario, the same data object may need to be published in multiple different languages in order to be available for users in multiple different language countries or regions to browse. At this time, if the label is still automatically generated based on the text feature comparison, label generation needs to be performed for text features of different languages, respectively, and thus the workload is very large.

Therefore, how to automatically generate tags for data objects more efficiently becomes a technical problem to be solved by those skilled in the art.

Disclosure of Invention

The application provides a data object label generation method, a data object searching method, a data object label generation device and electronic equipment, and a more effective training sample can be obtained, so that a label can be generated for a data object more effectively and automatically.

The application provides the following scheme:

a method of generating a data object tag, comprising:

obtaining a classification model, wherein the classification model is obtained by training a training sample, the training sample comprises an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record comprises: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

determining a data object to be predicted and image information corresponding to the data object;

and inputting the image information of the data object to be predicted into the classification model to obtain a corresponding text label.

A method of providing data object search information, comprising:

establishing a data object information base, wherein text label information related to a data object is stored in the information base, and the text label information comprises a target vocabulary with a search hotspot attribute; the text label information is determined by predicting the image of the data object by using a classification model; the classification model is obtained by training a training sample, the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

receiving a search request and determining a search keyword;

and providing a search result according to the matching degree of the text label of the data object and the search keyword.

A method of obtaining training sample data, comprising:

determining a target vocabulary with a search hotspot attribute according to a user search record in a target time period, wherein the target vocabulary with the search hotspot attribute comprises: searching the vocabulary with the frequency meeting the target condition;

analyzing the user behavior record after the target vocabulary is used as the key word to initiate the search, and determining a target data object having a target relation with the target vocabulary from the search result;

and labeling the image corresponding to the target data object by using the target vocabulary, and determining the image with labeling information as a training sample, wherein the training sample is used for training a target model, and the target model is used for outputting the matched target vocabulary by taking the image of the data object to be predicted as input so as to be used for determining the text label of the data object to be predicted.

A method of processing a classification model, comprising:

acquiring a training sample, wherein the training sample comprises an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting according to a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record comprises: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

and training the classification model by using the training samples.

An apparatus for generating a data object tag, comprising:

a classification model obtaining unit, configured to obtain a classification model, where the classification model is obtained by training a training sample, the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

the image determining unit is used for determining a data object to be predicted and image information corresponding to the data object;

and the prediction unit is used for inputting the image information of the data object to be predicted into the classification model so as to obtain a corresponding text label.

An apparatus for providing data object search information, comprising:

the system comprises a data object information base establishing unit, a search hot spot searching unit and a database establishing unit, wherein the data object information base establishing unit is used for establishing a data object information base, text label information related to a data object is stored in the information base, and the text label information comprises a target vocabulary with a search hot spot attribute; the text label information is determined by predicting the image of the data object by using a classification model; the classification model is obtained by training a training sample, the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

a search request receiving unit for receiving a search request and determining a search keyword;

and the search result providing unit is used for providing a search result according to the matching degree of the text label of the data object and the search keyword.

An apparatus for acquiring training sample data, comprising:

the target vocabulary determining unit is used for determining target vocabularies with the search hotspot attributes according to user search records in a target time period, wherein the target vocabularies with the search hotspot attributes comprise: searching the vocabulary with the frequency meeting the target condition;

the target data object determining unit is used for analyzing the user behavior record after the target vocabulary is used as the key word for initiating the search, and determining a target data object having a target relation with the target vocabulary from the search result;

and the labeling unit is used for labeling the image corresponding to the target data object by using the target vocabulary, determining the image with labeling information as a training sample, wherein the training sample is used for training a target model, and the target model is used for outputting the matched target vocabulary by taking the image of the data object to be predicted as input so as to be used for determining a text label of the data object to be predicted.

A device for processing a classification model, comprising:

a sample obtaining unit, configured to obtain a training sample, where the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by performing statistics on a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

and the training unit is used for training the classification model by utilizing the training samples.

According to the specific embodiments provided herein, the present application discloses the following technical effects:

according to the embodiment of the application, in order to automatically generate the text label based on the image of the data object and enable the specific text label to have the attribute of the search hot spot, the text label can be processed in the stage of training the model. Specifically, a target vocabulary with a search hotspot attribute may be determined according to a user search record within a certain time period, and the target vocabulary may be added to the tag library. In addition, besides determining the target vocabulary, the data object image can be labeled by using the specific target vocabulary, and the labeled result can be used as a training sample to train the model. Therefore, the embodiment of the application can also analyze the user behavior records after searching by taking the target vocabulary as the keywords, and if the user executes the target behaviors on some data objects in the search records, the data objects can be proved to have deep association with the current search keywords. Furthermore, the target vocabulary corresponding to the search keyword can be used for labeling the image corresponding to the data object, and the data object image with the labeling information can be used as a training sample for training a specific model.

In the model training stage, when a specific training sample is acquired, the training sample is not directly read from data such as a labeling result of a merchant, but a data object which is associated with a specific search keyword in a certain depth is mined from the user behavior data by analyzing the user behavior after a user uses the target vocabulary as the keyword for searching, and an image of the data object is labeled by using the target vocabulary corresponding to the search keyword to serve as the training sample. By the method, the quality of the training sample can be improved, and the prediction accuracy of the model can be improved conveniently.

After the training of the model is completed, a target vocabulary matched with the image of the target data object can be determined from the tag library and added as a tag of the data object when the model is predicted. In the embodiment of the application, when the label is automatically generated, only the image of the data object is required to be used as input information, and the text content is not required to be relied on, so that the label can be used in a cross-border scene. In addition, because the target vocabulary in the specific tag library has the attribute of the search hot spot, and further the tag page added for the data object has the attribute of the search hot spot, the tag is more suitable for being used in a search scene, which is beneficial to improving the matching degree of the search result and the search requirement and simultaneously improving the exposure rate of the data object with the corresponding attribute.

Of course, it is not necessary for any product to achieve all of the above-described advantages at the same time for the practice of the present application.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.

FIG. 1 is a schematic diagram of a system architecture provided by an embodiment of the present application;

FIG. 2 is a flow chart of a first method provided by an embodiment of the present application;

FIG. 3 is a schematic diagram of a model structure provided by an embodiment of the present application;

FIG. 4 is a flow chart of a second method provided by embodiments of the present application;

FIG. 5 is a flow chart of a third method provided by embodiments of the present application;

FIG. 6 is a flow chart of a fourth method provided by embodiments of the present application;

FIG. 7 is a schematic diagram of a first apparatus provided by an embodiment of the present application;

FIG. 8 is a schematic diagram of a second apparatus provided by an embodiment of the present application;

FIG. 9 is a schematic diagram of a third apparatus provided by an embodiment of the present application;

FIG. 10 is a schematic diagram of a fourth apparatus provided by an embodiment of the present application;

FIG. 11 is a schematic diagram of an electronic device provided by an embodiment of the application; .

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments that can be derived from the embodiments given herein by a person of ordinary skill in the art are intended to be within the scope of the present disclosure.

In the embodiment of the application, for a cross-border scenario and the like, a scheme for automatically generating text labels of data objects based on images (for example, commodity maps, and the like) of the data objects may be provided. In a specific implementation, the model may be trained in advance using an image associated with a tag text, and after the training is completed, the model may be input using an image of the commodity object, and the model may automatically generate and output a text tag by performing processing such as feature extraction on the image.

Since the image of the data object is usually language-independent, for example, the specific data object is a commodity object corresponding to a certain piece of clothing, the image is usually a photograph of the clothing, or a photograph of the model in a state of wearing the clothing, etc., so as to convey the related information of the data object to the user through the specific image without depending on the text content. In other words, in a cross-border scenario, the text contents of the same data object in different languages may be different (each described in a different language), but the images thereof are often the same. Therefore, in the embodiment of the present application, the text label may be automatically generated based on the image associated with the data object.

However, the inventor of the present application also finds that, in the process of implementing the present application, there are generally two main application scenarios for automatically generating tags for data objects. One is to recommend tags to a user during the process of publishing data objects by a user such as a merchant, so that the user can select from the recommended tags. The other is to automatically generate tags for the published data objects so as to provide search results for the user through the matching degree of the tags and the search keywords in the process of searching the data objects and the like. Under two different scenes, the generated label types can have different emphasis points, so that the effectiveness of the labels under different scenes is improved. For example, for the first case, the generated tags may be primarily tags of the data object property class, e.g., the style, color, etc. of the data object belong to such tags. In the second case, since the search word is matched with the data object tag in the search scene, the generated tag may be a tag that is more easily matched with the search word in the search scene, so as to increase the exposure rate of the data object in the search scene.

For this reason, the embodiments of the present application also provide a corresponding solution for the second case. In the scheme, not only can the image of the data object be used as input information to automatically generate the label, but also the model can be trained by adopting a special training sample, so that the label generated by the model can have the properties of searching hot spot words and the like. Specifically, when the special training sample is obtained, first, search terms having a search hotspot attribute may be analyzed from search records of a plurality of users within a target time period (e.g., the last month, etc.). The so-called search hotspot attribute is the word that is used by the user as a search term more frequently than a threshold or ranked further up. Then, a data object image related to the hot vocabulary is acquired, and when the data object image is acquired, the data object image can be directly acquired from the data object with the label added in the data object information base. However, there may be instances where the data object tags already in the database are inaccurate, on the one hand, or, on the other hand. There are also a large number of data objects that are actually related to a hot word, but to which the tag is not added.

In view of the above situation, the embodiment of the present application further provides a scheme for obtaining a training sample according to a user behavior record in a specific search scenario. Specifically, after the hot word is determined, the user behavior information after the search is initiated with the hot word may be analyzed, for example, if a user clicks on a data object in the search result to check details, collect, add a to-be-purchased set (e.g., "shopping cart" or the like), purchase, or the like, it is proved that the data object and the current hot word have a deep binding relationship, and then a "deep binding" relationship may be established between the hot word and an image of the data object, and the hot word having the "deep binding relationship" and the image of the corresponding data object are used as training samples to train a specific model. In this way, the trained model learns the internal relationship between the image of the data object and the hot word, so that the image of the data object can be input into the model when the label is automatically generated for the data object through the model, and accordingly, the model can classify the image of the data object and determine a matched label, and the label is a label with the attribute of searching the hot word.

In a specific implementation, from the perspective of a system architecture, the embodiment of the present application may provide corresponding functions in a data object information service system, where the data object information service system may include a server and a client. The operations of collecting training samples, training models, automatically generating labels with the search hotspot attributes for the data objects and the like can be completed at the server. For example, the server may use the full amount of data objects (or a part of the data objects) in the system database as the data objects to be predicted, so as to generate tags with the attribute of the search hotspot for specific data objects respectively (of course, for data objects unrelated to the search hotspot, the generated tags may be null). Specifically, the predicted tags with the search hotspot attributes for the data objects can be stored in a data object information base, so that the exposure rate of the corresponding data objects is improved in a search scene. In addition, because there are often newly released data objects in the system, the server may further re-execute the operation of automatic tag generation at a certain frequency, and the re-executed automatic tag generation process is mainly performed on the newly released data objects in the system, and updates the data object information base. Specifically, the update may be performed once a day, or may be performed at other frequencies, for example, every two days or every week, and so on. In addition, because the search hot spot attribute of the specific vocabulary may also be dynamically changed, in practical application, the training work of the specific model may also be performed again at intervals, when the model training is performed again, the target vocabulary with the search hot spot attribute may be determined again, the image of the data object may be marked again according to the behavior record of the user on the search result, and the like.

The client is mainly used for interacting with the user, for example, a search entry may be provided in a specific page, the user may input a specific keyword to initiate a search, and at this time, the server may also provide a specific search result based on the matching degree between the tag of the specific data object in the data object information base and the keyword.

The following describes in detail specific implementations provided in embodiments of the present application.

Example one

First, a method for obtaining training sample data is provided in the first embodiment, where an execution subject of the first embodiment may be the foregoing server, and specifically, referring to fig. 2, the method may include:

s201: determining a target vocabulary with a search hotspot attribute according to a user search record in a target time period, wherein the target vocabulary with the search hotspot attribute comprises: searching the vocabulary with the frequency meeting the target condition;

in the embodiment of the application, the exposure rate of the data object in the search scene is improved mainly by adding the tag with the search hotspot attribute to the data object. The search hotspot attribute is usually a dynamically changing attribute, for example, a word may have a high search heat in a certain period of time, the heat may decrease in the period of time, and so on. There are various factors in which the degree of heat of a specific search keyword varies with time, for example, some keywords may be affected by season, holidays, etc., some keywords may be affected by hot events, etc.

Thus, in particular implementations, it may be determined first which words may exist as tags with the attribute of a search hotspot. For this reason, in the embodiment of the present application, the determination may be made based on a user search record over a period of time. For example, according to the user search records in the last month, the search keywords with higher search times or higher frequency or higher rank are determined, and the words with the search hotspot attributes are determined according to the keywords. Such vocabulary can be added to the tag library, and after the training of a specific model is completed, specific tags are selected for specific data objects from the vocabulary.

When determining the target vocabulary with the search hotspot attribute, the occurrence frequency and the like of the search keywords can be directly counted, and then the search keywords meeting the conditions (for example, the search frequency is higher than the target threshold value) are determined as the target vocabulary according to the counting result. Or, the search keyword may be processed, and then the specific target vocabulary may be determined.

For example, if the search keyword language is not uniform due to the user being located in different countries, the search keyword may also be translated to a uniform language (such as chinese or english) first.

In addition, word segmentation processing can be carried out on the search keywords, and words without practical meaning contained in the search keywords are filtered out. For example, in the case of english word segmentation, since the basic unit of english language is a word, the word can be first partitioned by spaces, symbols, paragraphs, and the like, and a partition operation can be performed using a regular expression to segment a word group of a single search record into a plurality of words. And then, high-frequency words such as and/of/the/a and the like are excluded, and the high-frequency words can bring great influence on word frequency statistics and have no actual hot word meaning.

In addition, since there are some temporal and singular/plural labels in the english language, it is also possible to reduce the plural number of words to singular and reduce the temporal changes such as ing, ed, etc. to the normal current time. After word segmentation, filtering, word stem extraction and other processing, the occurrence frequency of words and the like are counted to determine the words with the search hotspot attribute. For example, the word frequency after word segmentation may be counted according to the number of search entries, and a certain frequency threshold is selected, and a high-frequency hotword is defined above the frequency threshold, and so on.

S202: analyzing the user behavior record after the target vocabulary is used as the key word to initiate the search, and determining a target data object having a target relation with the target vocabulary from the search result;

after the vocabulary with the search hotspot attribute is acquired, the user behavior record after searching by taking the target vocabulary as the keyword can be analyzed. Specifically, since the search result list page may generally include a plurality of search results, the user may acquire summary information about each search result, including pictures, titles, price attribute information, and the like, from the list page, and may determine whether to click on a specific search result to enter a corresponding data object detail page for browsing based on the summary information, and after entering the detail page, may also determine whether to perform further collection, join a to-be-purchased collection (e.g., a shopping cart, and the like), and acquire customer service, create an order, a deal, and the like provided by the user of the seller in a manner of "private chat" and the like according to the detail information.

If the user performs the above-mentioned actions on a certain data object, the data object has a strong association degree with the search keyword, or the search keyword can be used as a description of the data objects. Accordingly, it may be determined that the data object has the target relationship with the search keyword, which may also be referred to as a "strong association relationship," or the like.

S203: and labeling the image corresponding to the target data object by using the target vocabulary, and determining the image with labeling information as a training sample, wherein the training sample is used for training a target model, and the target model is used for outputting the matched target vocabulary by taking the image of the data object to be predicted as input so as to be used for determining the text label of the data object to be predicted.

After the data object having the target relationship with the target vocabulary is determined, the association relationship between the image corresponding to such data object and the target vocabulary may be determined as a training sample, and then the target model is trained. That is, since the tag is automatically generated based on the image of the data object in the embodiment of the present application, after the data object and the target vocabulary having the binding relationship are specifically determined, the image of the data object (for example, the main graph of the commodity object) may be labeled by using the target vocabulary, and the data object image with the labeling information may be used as a training sample for training a specific model.

It should be noted that the types of the target words are various, and in the foregoing manner, the image of each data object may be labeled with a plurality of target words. In the embodiment of the application, the image of the data object is mainly analyzed from a visual angle, so that some target vocabulary labels which cannot be visually recognized can be screened out during specific implementation, and the labels related to the image of the data object are visually-determinable labels. For example, some brand vocabularies, or vocabularies of sales attribute class such as "package mail", etc. may be filtered out to improve the quality of the training sample.

The visually separable target vocabularies screened according to the above steps can be added into a tag library, and each tag in the tag library can be respectively subjected to digital coding, so that each tag corresponds to a unique digital identifier, and further, coding association of multiple tags can be completed for each image. And then entering a specific model training process.

For a specific model, a plurality of specific algorithms or structures can be adopted for implementation, for example, in one mode, a multi-label neural network can be constructed. An alternative network structure can be shown in fig. 3, wherein "circles" in a column on the left represent a backbone network, and particularly, ResNet50 can be used as a backbone neural network (not limited to this network structure); the right column of the circle part represents the neuron in the last layer, and particularly, a sigmoid function can be adopted as an activation function of an output layer, and a cross entropy loss function can be selected, and the like.

After the structure of the model and the like are determined, the model can be trained by using the previously acquired training samples, and through multiple iterations, parameters in the model (for example, a weight matrix in a neural network and the like) can be gradually updated. After training is completed, a specific model may be used to generate specific text labels for the data objects. The text tag is a word with a search hotspot attribute, so that a search result can be provided for a user subsequently in a search scene by using the text tag, so that the matching degree of the search result and a search requirement is improved, and meanwhile, the exposure rate of a data object with the corresponding attribute is also improved.

In summary, according to the embodiment of the present application, in order to automatically generate a text tag based on an image of a data object and enable a specific text tag to have a search hotspot attribute, processing may be performed at a stage of training a model. Specifically, a target vocabulary with a search hotspot attribute may be determined according to a user search record within a certain time period, and the target vocabulary may be added to the tag library. In addition, besides determining the target vocabulary, the data object image can be labeled by using the specific target vocabulary, and the labeled result can be used as a training sample to train the model. Therefore, the embodiment of the application can also analyze the user behavior records after searching by taking the target vocabulary as the keywords, and if the user executes the target behaviors on some data objects in the search records, the data objects can be proved to have deep association with the current search keywords. Furthermore, the target vocabulary corresponding to the search keyword can be used for labeling the image corresponding to the data object, and the data object image with the labeling information can be used as a training sample for training a specific model. In the model training stage, when a specific training sample is acquired, the training sample is not directly read from data such as a labeling result of a merchant, but a data object which is associated with a specific search keyword in a certain depth is mined from the user behavior data by analyzing the user behavior after a user uses the target vocabulary as the keyword for searching, and an image of the data object is labeled by using the target vocabulary corresponding to the search keyword to serve as the training sample. By the method, the quality of the training sample can be improved, and the prediction accuracy of the model can be improved conveniently.

Example two

The second embodiment corresponds to the first embodiment, and provides a method for processing a classification model for a model training process after a training sample is obtained, with reference to fig. 4, the method may include:

s401: acquiring a training sample, wherein the training sample comprises an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting according to a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record comprises: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

s402: and training the classification model by using the training samples.

EXAMPLE III

A third embodiment is also corresponding to the first embodiment, and for the process of predicting the data object by using the classification model after completing the model training, a method for generating a data object label is provided, referring to fig. 5, and the method may include:

s501: obtaining a classification model, wherein the classification model is obtained by training a training sample, the training sample comprises an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record comprises: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

and if the user performs a target operation on one or more data objects in the search result, the one or more data objects are the target data objects. The target operation may specifically include: browsing a detail page, collecting, joining a to-be-purchased collection, creating an order, making a deal, or otherwise obtaining customer service resources.

S502: determining a data object to be predicted and image information corresponding to the data object;

s503: and inputting the image information of the data object to be predicted into the classification model to obtain a corresponding text label.

Example four

The fourth embodiment also corresponds to the first embodiment, and provides a method for providing data object search information for a specific data object search process, referring to fig. 6, where the method may specifically include:

s601: establishing a data object information base, wherein text label information related to a data object is stored in the information base, and the text label information comprises a target vocabulary with a search hotspot attribute; the text label information is determined by predicting the image of the data object by using a classification model; the classification model is obtained by training a training sample, the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

S602: receiving a search request and determining a search keyword;

s603: and providing a search result according to the matching degree of the text label of the data object and the search keyword.

In concrete implementation, the visual semantic information of the data object image related to the search result can be obtained, and the search result is grouped according to the visual semantic information and then returned to the client for display. The visual semantics refers to the result of gradual accumulation of non-language semantic information, such as type color information in an image, or the movement characteristics of an animal or a person in a dynamic image. The visual semantics of a specific image can be obtained in various ways, for example, extraction can be performed by using a correlation model in the prior art, and the like. In this way, search results with the same visual semantics can be aggregated together and provided for the user, and the user can browse conveniently.

For the parts of the second to fourth embodiments that are not described in detail, reference may be made to the description of the first embodiment, which is not repeated herein.

It should be noted that, in the embodiments of the present application, the user data may be used, and in practical applications, the user-specific personal data may be used in the scheme described herein within the scope permitted by the applicable law, under the condition of meeting the requirements of the applicable law and regulations in the country (for example, the user explicitly agrees, the user is informed, etc.).

Corresponding to the first embodiment, an embodiment of the present application further provides an apparatus for acquiring training sample data, and referring to fig. 7, the apparatus may include:

a target vocabulary determining unit 701, configured to determine, according to a user search record in a target time period, a target vocabulary with a search hotspot attribute, where the target vocabulary with the search hotspot attribute includes: searching the vocabulary with the frequency meeting the target condition;

a target data object determining unit 702, configured to analyze a user behavior record after a search is initiated with the target vocabulary as a keyword, and determine a target data object having a target relationship with the target vocabulary from a search result;

the labeling unit 703 is configured to label an image corresponding to the target data object with the target vocabulary, and determine the image with labeling information as a training sample, where the training sample is used to train a target model, and the target model is used to output a matched target vocabulary by using the image of the data object to be predicted as an input, so as to determine a text label of the data object to be predicted.

The target data object determination unit may be specifically configured to:

and if the user performs a target operation on one or more data objects in the search result, determining the one or more data objects as the target data objects.

Wherein the target operation comprises: browsing a detail page, collecting, joining a to-be-purchased collection, creating an order, making a deal, or otherwise obtaining customer service resources.

Specifically, the user search records in the target time period include search keywords corresponding to a plurality of different languages;

the target vocabulary determining unit may specifically include:

the translation subunit is used for translating the search keywords in the plurality of different languages into a target language;

the word segmentation processing subunit is used for carrying out word segmentation processing on the search keywords under the target language to obtain a plurality of words;

the irrelevant word filtering subunit is used for filtering out the words irrelevant to the description information of the data object and then counting the searching frequency information of the words;

and the target vocabulary determining subunit is used for determining the target vocabulary with the search hotspot attribute according to the search frequency information of each vocabulary.

Wherein if the target language is english, the apparatus may further include:

the word stem extraction unit is used for carrying out word stem extraction after a plurality of words are obtained through word segmentation processing, wherein the word stem extraction comprises the following steps: the plural states of the vocabulary are reduced to the singular, or the tense changes of the vocabulary are reduced to the common present time.

In addition, the apparatus may further include:

and the vocabulary filtering unit is used for filtering out target vocabularies which cannot be identified by computer vision.

And the digital coding unit is used for carrying out digital coding on the target vocabulary and carrying out corresponding digital coding association on the image of the target data object so as to train the target model.

Corresponding to the second embodiment, an embodiment of the present application further provides a device for processing a classification model, and referring to fig. 8, the device may include:

a sample obtaining unit 801, configured to obtain a training sample, where the training sample includes an image of a target data object with tagging information, where the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by performing statistics on a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

a training unit 802, configured to train the classification model by using the training samples.

Corresponding to the embodiment, the embodiment of the present application further provides an apparatus for generating a data object tag, and referring to fig. 9, the apparatus may include:

a classification model obtaining unit 901, configured to obtain a classification model, where the classification model is obtained by training a training sample, where the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by performing statistics on a user search record in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

an image determining unit 902, configured to determine a data object to be predicted and image information corresponding to the data object;

a prediction unit 903, configured to input image information of the data object to be predicted into the classification model to obtain a corresponding text label.

Corresponding to the fourth embodiment, an embodiment of the present application further provides an apparatus for providing data object search information, and referring to fig. 10, the apparatus may include:

a data object information base establishing unit 1001, configured to establish a data object information base, where text tag information associated with a data object is stored in the information base, and the text tag information includes a target vocabulary with a search hotspot attribute; the text label information is determined by predicting the image of the data object by using a classification model; the classification model is obtained by training a training sample, the training sample includes an image of a target data object with tagging information, the tagging information is a target vocabulary with a search hotspot attribute, the target vocabulary is determined by counting user search records in a target time period, a tagging result of the image of the target data object is determined by analyzing a target user behavior record, and the target user behavior record includes: initiating a user behavior record after searching by taking the target vocabulary as a keyword;

a search request receiving unit 1002, configured to receive a search request and determine a search keyword;

a search result providing unit 1003, configured to provide a search result according to a matching degree between the text label of the data object and the search keyword.

In a specific implementation, the apparatus may further include:

the visual semantic acquisition unit is used for acquiring visual semantic information of the data object image related to the search result;

and the grouping unit is used for grouping the search results according to the visual semantic information.

In addition, the present application also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps of the method described in any of the preceding method embodiments.

And an electronic device comprising:

one or more processors; and

a memory associated with the one or more processors for storing program instructions that, when read and executed by the one or more processors, perform the steps of the method of any of the preceding method embodiments.

FIG. 11 illustrates an architecture of an electronic device, which may include, in particular, a processor 1110, a video display adapter 1111, a disk drive 1112, an input/output interface 1113, a network interface 1114, and a memory 1120. The processor 1110, the video display adapter 1111, the disk drive 1112, the input/output interface 1113, the network interface 1114, and the memory 1120 may be communicatively connected by a communication bus 1130.

The processor 1110 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solution provided by the present Application.

The Memory 1120 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1120 may store an operating system 1121 for controlling the operation of the electronic device 1100, a Basic Input Output System (BIOS) for controlling low-level operations of the electronic device 1100. In addition, a web browser 1123, a data store management system 1124, and a tag processing system 1125, among others, may also be stored. The tag processing system 1125 may be an application program for implementing the operations of the foregoing steps in this embodiment. In summary, when the technical solution provided by the present application is implemented by software or firmware, the relevant program codes are stored in the memory 1120 and called for execution by the processor 1110.

The input/output interface 1113 is used for connecting an input/output module to realize information input and output. The i/o module may be configured as a component in a device (not shown) or may be external to the device to provide a corresponding function. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.

Network interface 1114 is used to connect to a communications module (not shown) to enable the device to interact with other devices for communication. The communication module can realize communication in a wired mode (such as USB, network cable and the like) and also can realize communication in a wireless mode (such as mobile network, WIFI, Bluetooth and the like).

Bus 1130 includes a path that transfers information between the various components of the device, such as processor 1110, video display adapter 1111, disk drive 1112, input/output interface 1113, network interface 1114, and memory 1120.

It should be noted that although the above devices only show the processor 1110, the video display adapter 1111, the disk drive 1112, the input/output interface 1113, the network interface 1114, the memory 1120, the bus 1130 and so on, in a specific implementation, the devices may also include other components necessary for normal operation. Furthermore, it will be understood by those skilled in the art that the apparatus described above may also include only the components necessary to implement the solution of the present application, and not necessarily all of the components shown in the figures.

From the above description of the embodiments, it is clear to those skilled in the art that the present application can be implemented by software plus necessary general hardware platform. Based on such understanding, the technical solutions of the present application may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments of the present application.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the system or system embodiments are substantially similar to the method embodiments and therefore are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for related points. The above-described system and system embodiments are only illustrative, wherein the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

The method, the device and the electronic device for generating the data object tag and searching the data object provided by the present application are introduced in detail, and a specific example is applied in the present application to explain the principle and the implementation of the present application, and the description of the above embodiment is only used to help understanding the method and the core idea of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, the specific embodiments and the application range may be changed. In view of the above, the description should not be taken as limiting the application.

Claims

1. A method of generating a data object tag, comprising:

2. The method of claim 1,

and if the user performs a target operation on one or more data objects in the search result, the one or more data objects are the target data objects.

3. The method of claim 2,

the target operation comprises: browsing a detail page, collecting, joining a to-be-purchased collection, creating an order, making a deal, or otherwise obtaining customer service resources.

4. A method for providing data object search information, comprising:

receiving a search request and determining a search keyword;

5. The method of claim 4,

6. The method of claim 5,

7. The method of claim 4, further comprising:

acquiring visual semantic information of a data object image associated with the search result;

and grouping the search results according to the visual semantic information.

8. A method for obtaining training sample data, comprising:

9. The method of claim 8,

the determining, from the search results, a target data object having a target relationship with the hotspot vocabulary includes:

10. The method of claim 9,

11. The method of claim 8,

the user search records in the target time period comprise search keywords corresponding to a plurality of different languages;

the determining a target vocabulary with a search hotspot attribute according to the user search record in the target time period comprises:

translating the search keywords of the plurality of different languages into a target language;

performing word segmentation processing on the search keywords under the target language to obtain a plurality of words;

after filtering out the vocabulary irrelevant to the description information of the data object, counting the search frequency information of the vocabulary;

and determining a target vocabulary with the attribute of the search hotspot according to the search frequency information of each vocabulary.

12. The method of claim 11,

if the target language is English, the method further comprises:

performing stem extraction after obtaining a plurality of vocabularies through word segmentation processing, wherein the stem extraction comprises the following steps: the plural states of the vocabulary are reduced to the singular, or the tense changes of the vocabulary are reduced to the common present time.

13. The method of claim 8, further comprising:

and filtering out target vocabularies which cannot be identified by computer vision.

14. The method of claim 8, further comprising:

and carrying out digital coding on the target vocabulary, and carrying out corresponding digital coding association on the image of the target data object so as to train the target model.

15. A method for processing a classification model, comprising:

and training the classification model by using the training samples.

16. An apparatus for generating a data object tag, comprising:

17. An apparatus for providing data object search information, comprising:

18. An apparatus for acquiring training sample data, comprising:

19. An apparatus for processing a classification model, comprising:

20. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 15.

21. An electronic device, comprising:

one or more processors; and

a memory associated with the one or more processors for storing program instructions that, when read and executed by the one or more processors, perform the steps of the method of any of claims 1 to 15.