Search result determination method and device, electronic equipment and storage medium

Through the search result category determination model and affiliation filtering technology, the accuracy and relevance of search results are improved, the problem of insufficient accuracy of search results in the existing technology is solved, and the user experience is improved.

CN118885656APending Publication Date: 2024-11-01BEIJING ZITIAO NETWORK TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410910673.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-08
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

Existing search technology is difficult to improve the accuracy of search results and affects the user's search experience.

Method used

The model is determined through the search result category, and the target search result category is determined based on the semantic characteristics of the search text entered by the user, and the relevant search results are filtered according to the affiliation relationship, and the semantic information is associated to the target search results related to the user's search text.

Benefits of technology

It improves the accuracy and relevance of search results, improves the user's search experience, and avoids the problem of irrelevance of semantic information in the prior art.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118885656A_ABST
    Figure CN118885656A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a search result determination method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a search text of a user, determining a model through a search result category, and determining the search result category of the user according to semantic features of the search text; determining a target search result category corresponding to the search text in each predetermined first search result category; determining a first search result belonging to the target search result category according to the target search result category and a predetermined affiliation relationship between the first search result category and the first search result; wherein the semantic information of the first search results belonging to the same first search result category is associated; and determining a target search result corresponding to the search text according to the first search result belonging to the target search result category.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a search result determination method, device, electronic device, and storage medium. Background Art

[0002] Currently, users can search for the information they need through the Internet, for example, through search engines, or in business platforms such as short video platforms. For a search keyword entered by a user, multiple search results can often be determined on the Internet, and the accuracy of each search result directly affects the user's search experience. Therefore, it is necessary to provide a technical solution to improve the accuracy of the determined search results and improve the user's search experience. Summary of the invention

[0003] The embodiments of the present disclosure provide a search result determination method, device, electronic device, and storage medium, which can improve the accuracy of the determined search results and enhance the user's search experience.

[0004] In a first aspect, an embodiment of the present disclosure provides a method for determining a search result, including:

[0005] Obtaining a user's search text, and determining a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0006] Determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0007] The target search result corresponding to the search text is determined according to the first search result belonging to the target search result category.

[0008] In a second aspect, an embodiment of the present disclosure provides a search result determination device, including:

[0009] A category determination unit, configured to obtain a user's search text, and determine a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0010] A first result determination unit, configured to determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0011] The second result determination unit is configured to determine a target search result corresponding to the search text according to the first search result belonging to the target search result category.

[0012] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor; and a memory configured to store computer-executable instructions, wherein the computer-executable instructions, when executed, enable the processor to implement the steps of the method described in the first aspect above.

[0013] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store computer-executable instructions, and the computer-executable instructions implement the steps of the method described in the first aspect when executed by a processor.

[0014] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0015] In one or more embodiments of the present disclosure, first, the user's search text is obtained, and a model is determined through the search result category. According to the semantic features of the search text, the target search result category corresponding to the search text is determined in each predetermined first search result category. Then, according to the target search result category and the subordinate relationship between the predetermined first search result category and the first search result, the first search result belonging to the target search result category is determined, wherein the semantic information of the first search results belonging to the same first search result category is associated. Finally, according to the first search result belonging to the target search result category, the target search result corresponding to the search text is determined. It can be seen that through this embodiment, since the model can be determined through the search result category, the target search result category corresponding to the search text can be determined according to the semantic features of the search text, and the target search result corresponding to the search text can be determined according to the first search result belonging to the target search result category, the target search result can be associated with the semantic information of the search text, and since the first search results associated with the semantic information are pre-classified into the same first search result category, the semantic information of each first search result belonging to the target search result category is also associated with each other, and therefore the semantic information of each target search result is also associated with each other. It can be seen that through this embodiment, target search results associated with the semantic information of the user's search text can be provided to the user, and the semantic information of each target search result is also associated with each other, thereby improving the accuracy of the determined target search results and improving the user's search experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate one or more embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

[0017] Figure 1 A flowchart of a method for determining search results provided by an embodiment of the present disclosure;

[0018] Figure 2 A schematic diagram of an application of a search result category determination model provided by an embodiment of the present disclosure;

[0019] Figure 3 A schematic diagram of a flow chart of establishing a subordinate relationship between a first search result category and a first search result provided in an embodiment of the present disclosure;

[0020] Figure 4 A schematic diagram of a training process of a search result category determination model provided in an embodiment of the present disclosure;

[0021] Figure 5 A schematic diagram of training a search result category determination model provided in an embodiment of the present disclosure;

[0022] Figure 6 A schematic diagram of the training process of a semantic extraction model provided in one embodiment of the present disclosure;

[0023] Figure 7 A schematic diagram of training a second neural network provided in one embodiment of the present disclosure;

[0024] Figure 8 A flowchart of a search result determination method provided by another embodiment of the present disclosure;

[0025] Fig. 9 A schematic diagram of the structure of a search result determination device provided by an embodiment of the present disclosure;

[0026] Fig.10 A schematic diagram of the structure of an electronic device provided in one embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of the present disclosure, the technical solutions in one or more embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in one or more embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on one or more embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the protection scope of the present disclosure.

[0028] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0029] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0030] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0031] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0032] The disclosed embodiment provides a method for determining search results, which can improve the accuracy of the determined search results and improve the user's search experience. The method can be applied to a backend and executed by the backend, which can be in the form of a server cluster or a single server, or in the form of a cloud server. The method can also be applied to a terminal device and executed by the terminal device, which includes but is not limited to mobile phones, computers, tablet computers, laptops, car computers, wearable devices and other terminal devices.

[0033] Figure 1 A flowchart of a method for determining search results provided by an embodiment of the present disclosure is shown in FIG. Figure 1 As shown, the process includes:

[0034] Step S102, obtaining the user's search text, and determining the target search result category corresponding to the search text in each predetermined first search result category through a search result category determination model according to the semantic features of the search text;

[0035] Step S104, determining the first search results subordinate to the target search result category according to the target search result category and the predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0036] Step S106: determining a target search result corresponding to the search text according to the first search result belonging to the target search result category.

[0037] In this embodiment, first, the user's search text is obtained, and a model is determined through the search result category determination model. According to the semantic features of the search text, the target search result category corresponding to the search text is determined in each predetermined first search result category. Then, according to the target search result category and the subordinate relationship between the predetermined first search result category and the first search result, the first search result belonging to the target search result category is determined, wherein the semantic information of the first search results belonging to the same first search result category is associated. Finally, according to the first search result belonging to the target search result category, the target search result corresponding to the search text is determined. It can be seen that through this embodiment, since the target search result category determination model can be used to determine the target search result category corresponding to the search text according to the semantic features of the search text, and the target search result corresponding to the search text can be determined according to the first search result belonging to the target search result category, the target search result can be associated with the semantic information of the search text, and since the first search results associated with the semantic information are pre-classified into the same first search result category, the semantic information of each first search result belonging to the target search result category is also associated with each other, and therefore the semantic information of each target search result is also associated with each other. It can be seen that through this embodiment, target search results associated with the semantic information of the user's search text can be provided to the user, and the semantic information of each target search result is also associated with each other, thereby improving the accuracy of the determined target search results and improving the user's search experience.

[0038] In the above step S102, the user's search text is obtained, and the user's search text can be any one of the search words and search terms entered by the user in the search box. In the above step S102, the search result category determination model is also used, and the target search result category corresponding to the search text is determined in each predetermined first search result category according to the semantic features of the search text. The semantic features of the search text can be represented by a matrix of size N*M, where N and M are both natural numbers. The semantic features of the search text are used to represent the semantic information of the search text. When the text characters included in the two search texts are different but the semantic information is similar, the semantic features of the two search texts are also similar. In this embodiment, multiple first search result categories are predetermined, and the first search result category can be represented by a positive integer, and the target search result category corresponding to the search text can also be represented by a positive integer. The target search result category can represent the category of the target search result corresponding to the user's search text, and the target search result category can be the first search result category where the target search result corresponding to the user's search text is located.

[0039] In one embodiment, the search result category determination model is used to determine the target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text, including:

[0040] Determine the first feature extraction network in the model through the search result category to extract the semantic features of the search text;

[0041] Determine the first classification score for classifying the search text into each first search result category according to the semantic features of the search text through the first score determination network in the search result category determination model;

[0042] According to the first classification score, a target search result category corresponding to the search text is determined in each first search result category.

[0043] Figure 2 A schematic diagram of an application of a search result category determination model provided by an embodiment of the present disclosure, such as Figure 2 As shown, the search result category determination model includes a first feature extraction network and a first score determination network, the first feature extraction network includes a first feature extraction layer and a second feature extraction layer, the first feature extraction layer includes a Deberta Backbone network, and the second feature extraction layer includes a Projection Head. In this embodiment, the search text can be first input into the search result category determination model, and the semantic features of the search text can be extracted through the first feature extraction network in the search result category determination model. For example, the high-dimensional semantic features of the search text are extracted through the first feature extraction layer Deberta Backbone, and the high-dimensional semantic features of the search text are reduced in dimension through the second feature extraction layer Projection Head, and the reduced features are used as the semantic features of the search text, wherein the high-dimensional semantic features refer to semantic features with a large number of dimensions, and reducing the high-dimensional semantic features refers to reducing the number of dimensions of the semantic features.

[0044] Next, the first score determination network in the search result category determination model is used to determine the first classification score of the search text classified into each first search result category based on the semantic features of the search text. For example, the first classification score of the search text classified into the first first search result category is determined to be 20 points, and the first classification score of the search text classified into the second first search result category is determined to be 50 points. Finally, based on the first classification score of the search text classified into each first search result category, the target search result category corresponding to the search text is determined in each first search result category. The training process of the search result category determination model will be introduced later.

[0045] It can be seen that through this embodiment, the semantic features of the search text can be extracted through the first feature extraction network in the search result category determination model, and the first classification score for classifying the search text into each first search result category can be determined based on the semantic features of the search text through the first score determination network in the search result category determination model. According to the first classification score, the target search result category corresponding to the search text can be accurately determined in each first search result category, thereby improving the accuracy of determining the target search result category through classification.

[0046] In one embodiment, determining a first classification score for classifying the search text into each first search result category according to the semantic features of the search text includes:

[0047] Determining a first similarity between a semantic feature of the search text and a semantic feature of each first search result category; the semantic features of the first search result category are stored in a first score determination network;

[0048] A first classification score for classifying the search text into each first search result category is determined according to the first similarity.

[0049] In this embodiment, the first score determination network includes a linear layer, which has a code book. In the process of training the search result category determination model, the semantic features of each first search result category can be trained in the code book. Therefore, the semantic features of each first search result category can also be considered as part of the parameters of the first score determination network.

[0050] In this embodiment, the cosine distance between the semantic features of the search text and the semantic features of each first search result category is calculated through the first score determination network, and the inverse of the cosine distance is used as the first similarity between the semantic features of the search text and the semantic features of each first search result category. Next, the first classification score for classifying the search text into each first search result category is determined based on the first similarity. For example, the first similarity between the semantic features of the search text and the semantic features of each first search result category is used as the first classification score for classifying the search text into each first search result category. Therefore, for each first search result category, the higher the first similarity between the semantic features of the search text and the semantic features of the first search result category, the higher the first classification score for classifying the search text into the first search result category, and the more likely the search text is to be classified into the first search result category.

[0051] It can be seen that through this embodiment, the semantic features of each first search result category can be trained and stored in the first score determination network, so that the first similarity between the semantic features of the search text and the semantic features of each first search result category can be determined through the first score determination network, and the first classification score for classifying the search text into each first search result category is determined based on the first similarity, thereby improving the accuracy of determining the first classification score.

[0052] In one embodiment, determining the target search result category corresponding to the search text in each first search result category according to the first classification score includes:

[0053] Normalize the first classification score;

[0054] According to the normalized first classification score and the first classification score threshold, a target search result category corresponding to the search text is determined in each first search result category.

[0055] In this embodiment, after obtaining the first classification score of the search text classified into each first search result category, as shown in FIG. Figure 2 As shown, a normalization function such as a softmax function in the model is determined by the search result category to normalize each first classification score. Each normalized first classification score is a value between 0 and 1, and the sum of each normalized first classification score is 1.

[0056] Next, in each first search result category, a category whose normalized first classification score is greater than the first classification score threshold is determined, and the category is determined as the target search result category corresponding to the search text. That is, the target search result category is the category whose normalized first classification score is greater than the first classification score threshold in each first search result category, so that the user's search text is classified into the first search result category with a higher first classification score. In this embodiment, the number of target search result categories determined by the first classification score threshold can be one or more.

[0057] It can be seen that through this embodiment, the first classification score can be normalized, and the target search result category corresponding to the search text can be determined in each first search result category based on the normalized first classification score and the first classification score threshold, thereby improving the efficiency and accuracy of determining the target search result category, and one or more target search result categories can be determined to prepare for recalling more target search results.

[0058] In this embodiment, after the target search result category corresponding to the user's search text is determined, in the above step S104, the subordinate relationship between the predetermined first search result category and the first search result is also obtained.

[0059] In this embodiment, multiple first search results are obtained in advance. The first search result may be a search result stored in a database and provided to a user. For example, when a user performs a search operation in a search engine, the first search result is each search result stored in the database of the search engine. The first search result includes a page link and page data corresponding to the page link, and the page data includes at least one of video, audio, picture, and text.

[0060] In this embodiment, each first search result can be classified according to the semantic information of each first search result to obtain each first search result category, so that the semantic information of each first search result belonging to the same first search result category is associated, wherein each first search result category can be represented by a positive integer. In this embodiment, a subordinate relationship between the first search result category and the first search result is also established, and the subordinate relationship can be obtained in step S104.

[0061] Figure 3 A schematic diagram of a flow chart of establishing a subordinate relationship between a first search result category and a first search result provided in an embodiment of the present disclosure. In one embodiment, the following steps can be performed: Figure 3 According to the steps shown in , the first search results are classified to obtain various first search result categories, and a subordinate relationship between the first search result categories and the first search results is established. Figure 3 As shown, the process includes:

[0062] Step S302, obtaining a second identification text of the first search result; the length of the second identification text is greater than a second length threshold;

[0063] Step S304, extracting semantic features of the second identification text through a semantic extraction model;

[0064] Step S306: clustering the first search results according to the semantic features of the second identification text to obtain the subordinate relationship between the first search result category and the first search results.

[0065] In this embodiment, first, the second identification text of each first search result is obtained, and the length of the second identification text is greater than the second length threshold. In one embodiment, when the first search result includes documents or books, at least one of the title, introduction, and text of the first search result whose length is greater than the second length threshold is obtained as the second identification text, and when the first search result includes videos, audios, or pictures, at least one of the title, dialogue, and narration of the first search result whose length is greater than the second length threshold is obtained as the second identification text.

[0066] Next, the semantic features of each second identification text are extracted through the semantic extraction model. The semantic extraction model is a pre-trained model for extracting semantic features of a longer text. Each second identification text can be input into the semantic extraction model, and the semantic extraction model can output the semantic features of each second identification text. The training process of the semantic extraction model will be introduced later.

[0067] Finally, according to the semantic features of each second identification text, each first search result is clustered to obtain each first search result category, and a subordinate relationship between the first search result category and the first search result is established. For example, by using the Kmeans clustering method, according to the semantic features of each second identification text, each first search result is clustered to obtain each first search result category, and a subordinate relationship between the first search result category and the first search result is established.

[0068] It can be seen that through this embodiment, the semantic features of the second identification text of each first search result can be extracted, and the first search results can be clustered according to the semantic features of the second identification text to obtain the subordinate relationship between the first search result category and the first search result, so that the first search results associated with the semantic information belong to the same first search result category, thereby achieving the effect of creating each first search result category according to the relevance of the semantic information.

[0069] In the above step S104, after obtaining the subordinate relationship between the first search result category and the first search result, the first search results subordinate to the target search result category are determined according to the target search result category and the subordinate relationship between the first search result category and the first search result. For example, in the subordinate relationship between the first search result category and the first search result, the target search result category is searched, and the first search results subordinate to the target search result category are obtained from the subordinate relationship. The number of target search result categories can be one or more, and each target search result category can include multiple first search results.

[0070] In the above step S106, the target search result corresponding to the search text is determined based on the first search result belonging to the target search result category. For example, each first search result belonging to each target search result category is used as the target search result corresponding to the search text, or, among each first search result belonging to each target search result category, the target search result corresponding to the search text is obtained by filtering according to a certain filtering rule. The filtering rule may be related to at least one of the popularity information and the number of views of the first search result.

[0071] It can be understood that the semantic information of multiple first search results under a target search result category is associated with each other, and since each target search result category is determined based on the semantic features of the search text, the multiple first search results under each target search result category are associated with the semantic information of the search text, and therefore the semantic information of the multiple first search results under each target search result category is also associated with each other.

[0072] Therefore, in this embodiment, based on the user's search results, multiple first search results associated with the semantic information of the search text can be recalled for the user, and the semantic information between the recalled multiple first search results is also associated with each other, thereby achieving the recall of similar semantics, improving the accuracy of determining search results, and improving the user's search experience.

[0073] Furthermore, since when creating the subordinate relationship between the first search result category and the first search result, the subordinate relationship is obtained by clustering the first search result according to the semantic features of the second identification text of the first search result which is longer in length, and when determining the target search result, the corresponding target search result category is determined by the search result category determination model according to the search text input by the user, the ability to map the short text (search text) input by the user to the long text (search result) is realized through the search result category determination model, so that the semantically similar long text can be recalled according to the short text input by the user, thereby improving the recall ability of the search results. Compared with the method in the prior art of retrieving the long text vector of the search result similar to the short text vector of the short text input by the user in the database, and comparing the short text vector of the short text input by the user with the retrieved long text vector to recall the search results, this embodiment avoids the problem of insufficient retrieval result accuracy caused by retrieving the long text vector, avoids the loss of accuracy caused by the comparison of long and short vectors, improves the recall ability of the search results, and, compared with the scheme in the prior art of recommending search results to users according to the search results clicked by the user in history, this embodiment avoids the defect of directly recommending text-similar search results to users, and realizes the recall of search results based on semantic information.

[0074] The following describes the training process of the search result category determination model. Figure 4 A schematic diagram of a training process of a search result category determination model provided in an embodiment of the present disclosure, such as Figure 4 As shown, the process includes:

[0075] Step S402, obtaining a first identification text of the first search result; the length of the first identification text is less than a first length threshold;

[0076] Step S404, determining a corresponding relationship between the first search result category and the first identification text according to the subordinate relationship between the first search result category and the first search result;

[0077] Step S406, training a first neural network according to the correspondence between the first identification text, the first search result category and the first identification text; the trained first neural network is a search result category determination model.

[0078] In the above step S402, the first identification text of each first search result is obtained, and the length of the first identification text is less than the first length threshold. According to the previous introduction, the function of the search result category determination model is to determine the target search result category corresponding to the search text input by the user, that is, to achieve the ability to map the short text (search text) input by the user to the long text (search result). Therefore, when training the search result category determination model, it is necessary to select short text as training data. Therefore, the first identification text of the first search result is selected as training data here.

[0079] In the above step S404, the correspondence between the first search result category and the first identification text is determined based on the subordinate relationship between the first search result category and the first search result. For each first search result, the first identification text of the first search result corresponds to the first search result category to which the first search result belongs. For example, there are 10,000 first search results in total, and these 10,000 first search results are subordinate to 500 first search result categories. If a first identification text is obtained for each first search result, 10,000 first identification texts are obtained, and there is a correspondence between these 10,000 first identification texts and the above 500 first search result categories. Of course, in this embodiment, in order to reduce the amount of training data, the first identification text can be obtained for some but not all first search results.

[0080] In the above step S406, the first neural network is trained according to the correspondence between each first identification text, the first search result category and the first identification text, and the trained first neural network is a search result category determination model.

[0081] It can be seen that through this embodiment, the search result category determination model can be trained efficiently and accurately according to the correspondence between the first identification text of the first search result with a shorter length, the first search result category and the first identification text.

[0082] In one embodiment, obtaining the first identification text of the first search result includes:

[0083] At least one of the title, title keywords, and text keywords of the first search result is obtained as the first identification text of the first search result.

[0084] In this embodiment, at least one of the title, title keywords, and text keywords of the first search result whose length is less than the first length threshold can be obtained as the first identification text of the first search result, so as to train the search result category determination model using the first identification text.

[0085] In one embodiment, when the first search result includes documents or books, at least one of the title, title keywords, introduction keywords, and text keywords of the first search result whose length is less than a first length threshold is obtained as the first identification text; when the first search result includes videos or audio or pictures, at least one of the title, title keywords, dialogue keywords, and narration keywords of the first search result whose length is less than the first length threshold is obtained as the first identification text.

[0086] It can be seen that through this embodiment, a first identification text of shorter length can be obtained, and the search result category determination model can be trained using the first identification text, so that the search result category determination model learns the ability to map the short text (search text) input by the user to the long text (search result).

[0087] In one embodiment, training the first neural network according to the correspondence between the first identification text, the first search result category and the first identification text includes:

[0088] Extracting semantic features of the first identification text through a first feature extraction network in the first neural network;

[0089] Determine, by a first score determination network in the first neural network, a second classification score for classifying the first identification text into each first search result category according to the semantic features of the first identification text;

[0090] The first feature extraction network and the parameters in the first score determination network are trained according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text.

[0091] Figure 5 A schematic diagram of a training method for a search result category determination model provided in an embodiment of the present disclosure is shown in FIG. Figure 5As shown, the first neural network includes a first feature extraction network and a first score determination network, the first feature extraction network includes a first feature extraction layer and a second feature extraction layer, the first feature extraction layer includes a Deberta Backbone network, and the second feature extraction layer includes a Projection Head. In this embodiment, the first identification text can be first input into the first neural network, and the semantic features of each first identification text can be extracted through the first feature extraction network in the first neural network. For example, the high-dimensional semantic features of the first identification text are extracted through the first feature extraction layer Deberta Backbone, and the high-dimensional semantic features of the first identification text are reduced in dimension through the second feature extraction layer Projection Head, and the reduced features are used as the semantic features of the first identification text, wherein the high-dimensional semantic features refer to the semantic features with a large number of dimensions, and reducing the high-dimensional semantic features refers to reducing the number of dimensions of the semantic features.

[0092] Then, the first score determination network in the first neural network determines the second classification score of each first identification text classified into each first search result category according to the semantic features of each first identification text, for example, determining that the second classification score of a certain first identification text classified into the first first search result category is 20 points, and determining that the second classification score of a certain first identification text classified into the second first search result category is 50 points. It can be understood that the second classification score is obviously an inaccurate score obtained through learning.

[0093] Finally, the parameters in the first feature extraction network and the first score determination network are trained according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text. That is, with the classification of each first identification text into the corresponding first search result category as the training goal, the parameters in the first feature extraction network and the first score determination network are trained so that the approximate search result category corresponding to the first identification text determined according to the second classification score is the first search result category recorded in the correspondence between the first search result category and the first identification text.

[0094] It can be seen that through this embodiment, it is possible to classify each first identification text into the corresponding first search result category as a training goal, train the parameters in the first feature extraction network and the first score determination network, and improve the accuracy of the trained search result category determination model.

[0095] In one embodiment, determining, according to the semantic features of the first identification text, a second classification score for classifying the first identification text into each first search result category includes:

[0096] Determining a second similarity between the semantic feature of the first identification text and the semantic feature of each first search result category; the semantic feature of the first search result category is initialized in the first score determination network; the semantic feature of the first search result category is adjusted during the model training process;

[0097] A second classification score for classifying the first identification text into each first search result category is determined according to the second similarity.

[0098] In this embodiment, the first score determination network includes a linear layer having a code book. In the process of training the search result category determination model, the semantic features of each first search result category can be initialized in the code book. The semantic features of each first search result category are also part of the parameters of the first score determination network. The semantic features of each first search result category are adjusted during the model training process.

[0099] In this embodiment, for each first identification text, the cosine distance between the semantic features of the first identification text and the semantic features of each first search result category is calculated through the first score determination network, and the inverse of the cosine distance is used as the second similarity between the semantic features of the first identification text and the semantic features of each first search result category. Then, for each first identification text, the second classification score for classifying the first identification text into each first search result category is determined based on the second similarity. For example, the second similarity between the semantic features of the first identification text and the semantic features of each first search result category is used as the second classification score for classifying the first identification text into each first search result category. Therefore, for each first search result category and each first identification text, the higher the second similarity between the semantic features of the first identification text and the semantic features of the first search result category, the higher the second classification score for classifying the first identification text into the first search result category, and the more likely the first identification text is to be classified into the first search result category.

[0100] It can be seen that through this embodiment, the semantic features of each first search result category can be trained and stored in the first score determination network, so that the second similarity between the semantic features of the first identification text and the semantic features of each first search result category can be determined through the first score determination network, and the second classification score of the first identification text to each first search result category is determined according to the second similarity, thereby improving the accuracy of determining the second classification score.

[0101] In one embodiment, training the first feature extraction network and the parameters in the first score determination network according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text includes:

[0102] Normalize the second classification score;

[0103] Determining, in each first search result category, an approximate search result category corresponding to the first identification text according to the normalized second category score and the second category score threshold;

[0104] The parameters in the first feature extraction network and the first score determination network are trained according to the approximate search result category corresponding to the first identification text until the approximate search result category corresponding to the first identification text satisfies the correspondence between the first search result category and the first identification text.

[0105] In this embodiment, after obtaining the second classification score of each first identification text classified into each first search result category, as shown in FIG. Figure 5 As shown, each second classification score is normalized by a normalization function in the first neural network, such as a softmax function, and each normalized second classification score is a value of 0-1, and the sum of each normalized second classification score is 1.

[0106] Next, for each first identification text, in each first search result category, determine the category whose normalized second classification score is greater than the second classification score threshold, and determine the category as the approximate search result category corresponding to the first identification text. That is, for each first identification text, the approximate search result category corresponding to the first identification text is the category whose normalized second classification score is greater than the second classification score threshold in each first search result category. In one example, the first search result category with the largest normalized second classification score can also be used as the approximate search result category corresponding to the first identification text.

[0107] Finally, with the training goal of classifying each first identification text into the corresponding first search result category, the parameters in the first feature extraction network and the first score determination network are trained so that for each first identification text, the approximate search result category corresponding to the first identification text determined according to the second classification score is the first search result category recorded in the correspondence between the first search result category and the first identification text. Among them, the parameters in the first feature extraction network include parameters for extracting semantic features, and the parameters in the first score determination network include the semantic features of each of the above-mentioned first search result categories.

[0108] It can be seen that through this embodiment, the second classification score can be normalized, and the approximate search result category corresponding to the first identification text can be determined in each first search result category based on the normalized second classification score and the second classification score threshold. Based on the determination result, the parameters in the first feature extraction network and the first score determination network are trained with the classification of each first identification text to the corresponding first search result category as the training goal, and the search result category determination model is accurately trained.

[0109] The training process of the semantic extraction model is described below. According to the previous description, the semantic extraction model is used to extract the semantic features of the second identification text of the first search result. Since the length of the second identification text is greater than the second length threshold, the semantic extraction model is used to extract the semantic features of the long text.

[0110] Figure 6 A schematic diagram of the training process of the semantic extraction model provided in one embodiment of the present disclosure is shown in FIG. Figure 6 As shown, the process includes:

[0111] Step S602, obtaining a sample search result and a sample identification text of the sample search result; the length of the sample identification text is greater than a third length threshold;

[0112] Step S604: training a second neural network according to the sample identification text of the sample search results; the trained second neural network is used to extract sample semantic features of the sample identification text and classify the sample identification text of the same sample search results into the same sample search results according to the sample semantic features;

[0113] Step S606, generating a semantic extraction model according to the trained second neural network.

[0114] In this embodiment, a sample search result is first obtained, and the sample search result may be the first search result. In order to reduce the amount of training data, part of the first search result may be obtained as the sample search result. Also, a sample identification text of the sample search result is obtained, and the length of the sample identification text is greater than the third length threshold.

[0115] Next, the second neural network is trained based on the sample identification text of the sample search results. The trained second neural network can extract the sample semantic features of the sample identification text, and classify the sample identification text of the same sample search results into the same sample search results based on the sample semantic features. That is, the trained second neural network can classify the sample identification text of the same sample search results into the same sample search results based on the sample semantic features of the sample identification text.

[0116] Finally, a semantic extraction model is generated based on the trained second neural network. Since the function of the semantic extraction model is to extract semantic features, the part of the second neural network used for extracting semantic features can be used as the semantic extraction model.

[0117] Because when training the semantic extraction model, the semantic extraction model is determined according to the trained second neural network, and the trained second neural network can classify the sample identification texts of the same sample search result into the same sample search result according to the sample semantic features of the sample identification texts, therefore, when there are multiple sample identification texts corresponding to the same sample search result, the trained second neural network can represent the multiple sample identification texts through similar semantic features, so that the multiple sample identification texts are classified into the same sample search result. Therefore, the semantic extraction model determined according to the trained second neural network also has the ability to extract similar semantic features from texts with different text characters but similar semantics, that is, it has the ability to characterize similar semantics.

[0118] It can be seen that through this embodiment, the trained second neural network can extract the sample semantic features of the sample identification text and classify the sample identification text of the same sample search result into the same sample search result according to the sample semantic features, and generate a semantic extraction model according to the trained second neural network, so that the semantic extraction model can have the ability to extract similar semantic features from texts with different text characters but similar semantics, that is, the ability to characterize similar semantics, thereby improving the accuracy of semantic extraction by the semantic extraction model.

[0119] In one embodiment, obtaining a sample search result and a sample identification text of the sample search result includes:

[0120] Obtaining a sample search result and a third identification text of the sample search result;

[0121] Rewrite the third identification text based on the semantic information of the third identification text to obtain a fourth identification text;

[0122] A sample identification text is generated according to the third identification text and the fourth identification text.

[0123] In this embodiment, a sample search result is first obtained, and the sample search result may be the first search result. In order to reduce the amount of training data, part of the first search result may be obtained as the sample search result. Also, a sample identification text of the sample search result is obtained, and the length of the sample identification text is greater than the third length threshold.

[0124] In one embodiment, when the sample search results include documents or books, at least one item among the title, introduction, and text of the sample search results whose length is greater than a third length threshold is obtained as the third identification text; when the sample search results include videos, audios, or pictures, at least one item among the title, dialogue, and narration of the sample search results whose length is greater than the third length threshold is obtained as the third identification text.

[0125] Next, the third identification text is rewritten based on the semantic information of the third identification text to obtain a fourth identification text. Each third identification text can be rewritten based on the semantic information of the third identification text through a large language model, manual rewriting, or synonym replacement to obtain at least one fourth identification text corresponding to each third identification text. It should be noted that the semantic information of the third identification text and the corresponding fourth identification text is the same.

[0126] Finally, a sample identification text is generated according to the third identification text and the fourth identification text. For example, for a sample search result, each third identification text and each fourth identification text corresponding to the sample search result are used as the sample identification text corresponding to the sample search result. Therefore, a sample search result may correspond to multiple sample identification texts.

[0127] In one example, the same sample search result corresponds to multiple sample identification texts. For example, sample search results A and B, sample search result A corresponds to sample identification texts A1 and A2, and sample search result B corresponds to sample identification texts B1 and B2. If the text characters of A1 and B1 are similar, it means that the semantic information of sample search results A and B is similar, which makes them have sample identification texts with similar text characters. In this case, in the process of training the second neural network, since the text characters of A1 and B1 are similar, the second neural network will extract similar semantic features for A1 and B1, and since the sample semantic features of each sample identification text corresponding to each sample search result must also be similar (otherwise, it is impossible to classify each sample identification text corresponding to the same sample search result into the same sample search result), the second neural network will also extract semantic features similar to A1 for A2 and semantic features similar to B1 for B2. It can be seen that, on the one hand, for a sample search result, even if the text characters of the multiple sample identification texts corresponding to the sample search result are different, for example, A1 is different from A2, the sample semantic features of the multiple sample identification texts corresponding to the sample search result are also similar, so that the trained second neural network has the ability to extract similar semantic features from texts with different text characters but similar semantics, that is, it has the ability to characterize similar semantics; on the other hand, for multiple sample search results, when the semantic information of multiple sample search results is similar, even if the text characters of the sample identification texts corresponding to them are different, for example, A2 is different from B2, the sample semantic features of the sample identification texts corresponding to them are also similar, so that the trained second neural network can extract similar semantic features for sample search results with similar semantics, so that when clustering to obtain each first search result, the first search results with similar semantics can be clustered into one category.

[0128] It can be seen that through this embodiment, sample identification text can be generated by rewriting, so that the same sample search result can have multiple sample identification texts, so that the sample search results that originally have different text characters but similar semantics have the opportunity to have sample identification texts with similar text characters, thereby fully mining the semantic features of the sample search results. On the one hand, the semantic extraction model has the ability to extract similar semantic features from texts with different text characters but similar semantics. On the other hand, the semantic extraction model extracts similar semantic features for sample search results with similar semantics, so that when clustering to obtain each first search result, the first search results with similar semantics can be clustered into one category.

[0129] In one embodiment, training a second neural network according to the sample identification text of the sample search results includes:

[0130] Extracting sample semantic features of the sample identification text through a second feature extraction network in the second neural network;

[0131] Determining, by a second score determination network in a second neural network, a third classification score for classifying the sample identification text into each sample search result according to the sample semantic features of the sample identification text;

[0132] The second feature extraction network is trained according to the third classification score and the parameters in the second score determination network are determined until the third classification score indicates that the sample identification texts of the same sample search result are classified into the same sample search result.

[0133] Figure 7 A schematic diagram of training a second neural network provided in an embodiment of the present disclosure is shown in FIG. Figure 7 As shown, the second neural network includes a second feature extraction network and a second score determination network, the second feature extraction network includes a third feature extraction layer and a fourth feature extraction layer, the third feature extraction layer includes a Deberta Backbone network, and the fourth feature extraction layer includes a Projection Head. In this embodiment, each sample identification text can be first input into the second neural network, and the sample semantic features of each sample identification text can be extracted through the second feature extraction network in the second neural network. For example, the high-dimensional semantic features of the sample identification text are extracted through the third feature extraction layer Deberta Backbone, and the high-dimensional semantic features of the sample identification text are reduced in dimension through the fourth feature extraction layer Projection Head, and the reduced-dimensional features are used as the sample semantic features of the sample identification text, wherein the high-dimensional semantic features refer to semantic features with a large number of dimensions, and reducing the dimensionality of the high-dimensional semantic features refers to reducing the number of dimensions of the semantic features.

[0134] Then, the second score determination network in the second neural network determines the third classification score of each sample identification text classified into each sample search result according to the sample semantic features of each sample identification text, for example, the third classification score of a certain sample identification text classified into the first sample search result is determined to be 20 points, and the third classification score of a certain sample identification text classified into the second sample search result is determined to be 50 points. It can be understood that the third classification score is obviously an inaccurate score obtained through learning.

[0135] Finally, the parameters in the second feature extraction network and the second score determination network are trained according to the third classification score until the third classification score indicates that the sample identification text of the same sample search result is classified into the same sample search result. That is, the parameters in the second feature extraction network and the second score determination network are trained with the training goal of classifying the sample identification text of each sample search result into the corresponding sample search result.

[0136] In this embodiment, after the second neural network is trained, the second feature extraction network in the second neural network can be used as a semantic extraction model.

[0137] It can be seen that through this embodiment, the sample identification text of each sample search result can be classified into the corresponding sample search result as the training goal, and the parameters in the second feature extraction network and the second score determination network can be trained to improve the accuracy of the trained second neural network.

[0138] In one embodiment, determining the third classification score of the sample identification text classified into each sample search result according to the sample semantic feature of the sample identification text includes:

[0139] Determine a third similarity between the sample semantic feature of the sample identification text and the semantic feature of each sample search result; the second score determines the semantic feature of the sample search result initialized in the network; the semantic feature of the sample search result is adjusted during the model training process;

[0140] A third classification score for classifying the sample identification text into each sample search result is determined according to the third similarity.

[0141] In this embodiment, the second score determination network includes a linear layer, which has a code book. During the training of the second neural network, the semantic features of each sample search result can be initialized in the code book. The semantic features of each sample search result are also part of the parameters of the second score determination network. The semantic features of each sample search result are adjusted during the model training process.

[0142] In this embodiment, for each sample identification text, the cosine distance between the sample semantic feature of the sample identification text and the semantic feature of each sample search result is calculated through the second score determination network, and the inverse of the cosine distance is used as the third similarity between the sample semantic feature of the sample identification text and the semantic feature of each sample search result. Then, for each sample identification text, the third classification score for classifying the sample identification text to each sample search result is determined according to the third similarity. For example, the third similarity between the sample semantic feature of the sample identification text and the semantic feature of each sample search result is used as the third classification score for classifying the sample identification text to each sample search result. Therefore, for each sample search result and each sample identification text, the higher the third similarity between the sample semantic feature of the sample identification text and the semantic feature of the sample search result, the higher the third classification score for classifying the sample identification text to the sample search result, and the more likely the sample identification text is to be classified to the sample search result.

[0143] It can be seen that through this embodiment, the semantic features of each sample search result can be trained and stored in the second score determination network, so that the third similarity between the sample semantic features of the sample identification text and the semantic features of each sample search result can be determined through the second score determination network, and the third classification score of the sample identification text classification to each sample search result is determined according to the third similarity, thereby improving the accuracy of determining the third classification score.

[0144] In one embodiment, training the second feature extraction network and determining parameters in the second score network according to the third classification score until the third classification score indicates that the sample identification texts of the same sample search result are classified into the same sample search result includes:

[0145] Normalize the third category scores;

[0146] Determine, in each sample search result, an approximate search result corresponding to the sample identification text according to the normalized third classification score and the third classification score threshold;

[0147] The parameters in the second feature extraction network and the second score determination network are trained according to the approximate search results corresponding to the sample identification text, until the sample identification texts of the same sample search result are classified into the same sample search result.

[0148] In this embodiment, after obtaining the third classification score of each sample identification text classification to each sample search result, as shown in FIG. Figure 7 As shown, each third classification score is normalized by a normalization function in the second neural network, such as a softmax function, and each third classification score after normalization is a value of 0-1, and the sum of each normalized third classification score is 1.

[0149] Next, for each sample identification text, in each sample search result, determine the result whose normalized third classification score is greater than the third classification score threshold, and determine the result as the approximate search result corresponding to the sample identification text. That is, for each sample identification text, the approximate search result corresponding to the sample identification text is the result whose normalized third classification score is greater than the third classification score threshold in each sample search result. In one example, the sample search result with the largest normalized third classification score can also be used as the approximate search result corresponding to the sample identification text.

[0150] Finally, the training goal is to classify each sample identification text of the same sample search result into the same sample search result, and the corresponding relationship between the sample search result and the sample identification text is combined to train the parameters in the second feature extraction network and the second score determination network, so that the sample identification text of the same sample search result is classified into the same sample search result according to the third classification score, that is, the approximate search result corresponding to the sample identification text is the sample search result corresponding to the sample identification text. Among them, the parameters in the second feature extraction network include parameters for extracting semantic features, and the parameters in the second score determination network include the semantic features of each sample search result mentioned above.

[0151] It can be seen that through this embodiment, the third classification score can be normalized, and the approximate search results corresponding to the sample identification text can be determined in each sample search result based on the normalized third classification score and the third classification score threshold. Based on the determination result, the parameters in the second feature extraction network and the second score determination network are trained with the training goal of classifying each sample identification text of the same sample search result into the same sample search result, and the semantic extraction model is accurately trained.

[0152] Based on the above embodiments, Figure 8 A flowchart of a method for determining search results provided by another embodiment of the present disclosure is shown in FIG. Figure 8 As shown, the process includes:

[0153] Step S802, obtaining the first search text as the sample search text, rewriting the title and body of the sample search result, and using the title and body before and after the rewriting as sample identification text, wherein the length of the sample identification text is greater than a third length threshold;

[0154] Step S804: training a second neural network according to the sample identification text of the sample search results; the trained second neural network is used to extract sample semantic features of the sample identification text and classify the sample identification text of the same sample search results into the same sample search results according to the sample semantic features;

[0155] Step S806, using the second feature extraction network in the trained second neural network as a semantic extraction model;

[0156] Step S808, obtaining the body of the first search result as the second identification text, the length of the second identification text being greater than a second length threshold, and extracting semantic features of the second identification text through a semantic extraction model;

[0157] Step S810, clustering the first search results according to the semantic features of the second identification text to obtain various first search result categories, and obtaining the subordinate relationship between the first search result categories and the first search results;

[0158] Step S812, obtaining the title of the first search result as a first identification text, wherein the length of the first identification text is less than a first length threshold;

[0159] Step S814, determining a corresponding relationship between the first search result category and the first identification text according to the subordinate relationship between the first search result category and the first search result;

[0160] Step S816, training a first neural network according to the correspondence between the first identification text, the first search result category and the first identification text, the trained first neural network being a search result category determination model;

[0161] Step S818, obtaining the user's search text, and determining the target search result category corresponding to the search text in each first search result category through the search result category determination model according to the semantic features of the search text;

[0162] Step S820, determining a first search result subordinate to the target search result category according to the target search result category and the subordinate relationship between the first search result category and the first search result;

[0163] Step S822: determine the first search result belonging to the target search result category as the target search result corresponding to the search text.

[0164] In this embodiment, when a new first search result appears, the semantic features of the new first search result can be directly extracted through the above semantic extraction model, and the new first search result can be classified into the corresponding first search result category based on the semantic features.

[0165] Through the above embodiments, target search results associated with the semantic information of the user's search text can be provided to the user, and the semantic information of each target search result is also associated with each other, thereby improving the accuracy of the determined target search results and improving the user's search experience.

[0166] Fig. 9 A schematic diagram of the structure of a search result determination device provided by an embodiment of the present disclosure, such as Fig. 9 As shown, the device comprises:

[0167] A category determination unit 91 is used to obtain a user's search text, and determine a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0168] A first result determination unit 92 is used to determine the first search results belonging to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results belonging to the same first search result category is associated;

[0169] The second result determination unit 93 is configured to determine the target search result corresponding to the search text according to the first search result belonging to the target search result category.

[0170] Optionally, the category determination unit 91 is specifically configured to:

[0171] Extracting semantic features of the search text through a first feature extraction network in the search result category determination model;

[0172] Determine, by means of a first score determination network in the search result category determination model, a first classification score for classifying the search text into each of the first search result categories according to the semantic features of the search text;

[0173] According to the first classification score, a target search result category corresponding to the search text is determined in each of the first search result categories.

[0174] Optionally, the category determination unit 91 is further specifically configured to:

[0175] Determining a first similarity between a semantic feature of the search text and a semantic feature of each of the first search result categories; the semantic features of the first search result categories are stored in the first score determination network;

[0176] A first classification score for classifying the search text into each of the first search result categories is determined according to the first similarity.

[0177] Optionally, the category determination unit 91 is further specifically configured to:

[0178] Normalizing the first classification score;

[0179] A target search result category corresponding to the search text is determined in each of the first search result categories according to the normalized first category score and the first category score threshold.

[0180] Optionally, the method further includes a first training unit, configured to:

[0181] Acquire a first identification text of the first search result; the length of the first identification text is less than a first length threshold;

[0182] Determining a corresponding relationship between the first search result category and the first identification text according to a subordinate relationship between the first search result category and the first search result;

[0183] A first neural network is trained according to the correspondence between the first identification text, the first search result category and the first identification text; the trained first neural network determines a model for the search result category.

[0184] Optionally, the first training unit is specifically used for:

[0185] Extracting semantic features of the first identification text through a first feature extraction network in the first neural network;

[0186] Determining, by a first score determination network in the first neural network, a second classification score for classifying the first identification text into each of the first search result categories according to the semantic features of the first identification text;

[0187] The first feature extraction network and the parameters in the first score determination network are trained according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text.

[0188] Optionally, the first training unit is further specifically configured to:

[0189] Determining a second similarity between the semantic feature of the first identification text and the semantic feature of each of the first search result categories; the semantic features of the first search result category are initialized in the first score determination network; the semantic features of the first search result category are adjusted during the model training process;

[0190] Determine a second classification score for classifying the first identification text into each of the first search result categories according to the second similarity.

[0191] Optionally, the first training unit is further specifically configured to:

[0192] Normalizing the second classification score;

[0193] Determining, in each of the first search result categories, an approximate search result category corresponding to the first identification text according to the normalized second classification score and the second classification score threshold;

[0194] The parameters in the first feature extraction network and the first score determination network are trained according to the approximate search result category corresponding to the first identification text until the approximate search result category corresponding to the first identification text satisfies the correspondence between the first search result category and the first identification text.

[0195] Optionally, a clustering unit is further included, for:

[0196] Acquire a second identification text of the first search result; the length of the second identification text is greater than a second length threshold;

[0197] Extracting semantic features of the second identification text by using a semantic extraction model;

[0198] The first search results are clustered according to the semantic features of the second identification text to obtain the subordinate relationship between the first search result category and the first search result.

[0199] Optionally, the method further includes a second training unit, configured to:

[0200] Obtaining a sample search result and a sample identification text of the sample search result; the length of the sample identification text being greater than a third length threshold;

[0201] According to the sample identification text of the sample search results, a second neural network is trained; the trained second neural network is used to extract sample semantic features of the sample identification text and classify the sample identification text of the same sample search results into the same sample search results according to the sample semantic features;

[0202] The semantic extraction model is generated according to the trained second neural network.

[0203] Optionally, the second training unit is specifically used for:

[0204] Obtaining the sample search result and a third identification text of the sample search result;

[0205] Rewrite the third identification text based on the semantic information of the third identification text to obtain a fourth identification text;

[0206] The sample identification text is generated according to the third identification text and the fourth identification text.

[0207] Optionally, the second training unit is specifically used for:

[0208] Extracting sample semantic features of the sample identification text through a second feature extraction network in the second neural network;

[0209] Determining, by a second score determination network in the second neural network, a third classification score for classifying the sample identification text into each of the sample search results according to the sample semantic features of the sample identification text;

[0210] The second feature extraction network and the parameters in the second score determination network are trained according to the third classification score until the third classification score indicates that the sample identification text of the same sample search result is classified into the same sample search result.

[0211] Optionally, the second training unit is further specifically configured to:

[0212] Determining a third similarity between the sample semantic feature of the sample identification text and the semantic feature of each of the sample search results; the semantic features of the sample search results are initialized in the second score determination network; the semantic features of the sample search results are adjusted during the model training process;

[0213] A third classification score for classifying the sample identification text into each of the sample search results is determined according to the third similarity.

[0214] Optionally, the second training unit is further specifically configured to:

[0215] Normalizing the third classification score;

[0216] Determining, in each of the sample search results, an approximate search result corresponding to the sample identification text according to the normalized third classification score and the third classification score threshold;

[0217] The parameters in the second feature extraction network and the second score determination network are trained according to the approximate search results corresponding to the sample identification text until the sample identification text of the same sample search result is classified into the same sample search result.

[0218] The search result determination device in the embodiment of the present disclosure can implement each process of the above-mentioned search result determination method embodiment and achieve the same effect and function, which will not be repeated here.

[0219] An embodiment of the present disclosure further provides an electronic device, Fig.10 A schematic diagram of the structure of an electronic device provided by an embodiment of the present disclosure, such as Fig.10As shown, the electronic device may have relatively large differences due to different configurations or performances, and may include one or more processors 1001 and memory 1002, and one or more applications or data may be stored in the memory 1002. Among them, the memory 1002 may be a short-term storage or a persistent storage. The application stored in the memory 1002 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the electronic device. Furthermore, the processor 1001 may be configured to communicate with the memory 1002 and execute a series of computer executable instructions in the memory 1002 on the electronic device. The electronic device may also include one or more power supplies 1003, one or more wired or wireless network interfaces 1004, one or more input or output interfaces 1005, one or more keyboards 1006, etc.

[0220] In a specific embodiment, the electronic device includes a processor; and a memory configured to store computer executable instructions, wherein when the computer executable instructions are executed, the processor implements the following process:

[0221] Obtaining a user's search text, and determining a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0222] Determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0223] The target search result corresponding to the search text is determined according to the first search result belonging to the target search result category.

[0224] The electronic device in the embodiment of the present disclosure can implement each process of the above-mentioned search result determination method embodiment and achieve the same effect and function, which will not be repeated here.

[0225] Another embodiment of the present disclosure further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store computer-executable instructions, and when the computer-executable instructions are executed by a processor, the following process is implemented:

[0226] Obtaining a user's search text, and determining a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0227] Determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0228] The target search result corresponding to the search text is determined according to the first search result belonging to the target search result category.

[0229] The storage medium in the embodiment of the present disclosure can implement each process of the above-mentioned search result determination method embodiment and achieve the same effect and function, which will not be repeated here.

[0230] Another embodiment of the present disclosure further provides a computer program product, the computer program product comprising a computer program, and when the computer program is executed by a processor, the following process is implemented:

[0231] Obtaining a user's search text, and determining a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model;

[0232] Determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated;

[0233] The target search result corresponding to the search text is determined according to the first search result belonging to the target search result category.

[0234] The computer program product in the embodiment of the present disclosure can implement each process of the above-mentioned search result determination method embodiment and achieve the same effect and function, which will not be repeated here.

[0235] In various embodiments of the present disclosure, the computer-readable storage medium includes a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0236] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0237] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0238] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0239] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing the embodiments of the present disclosure, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0240] Those skilled in the art will appreciate that one or more embodiments of the present disclosure may be provided as a method, system or computer program product. Therefore, one or more embodiments of the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0241] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0242] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0243] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0244] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0245] One or more embodiments of the present disclosure may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present disclosure may also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0246] Each embodiment in the present disclosure is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0247] The above description is only an embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of the claims of the present disclosure.

Claims

1. A method for determining search results, characterized in that: include: Obtaining a user's search text, and determining a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model; Determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated; The target search result corresponding to the search text is determined according to the first search result belonging to the target search result category.

2. The method according to claim 1, characterized in that: The method of determining the target search result category corresponding to the search text in each predetermined first search result category by using the search result category determination model according to the semantic features of the search text includes: Extracting semantic features of the search text through a first feature extraction network in the search result category determination model; Determine, by means of a first score determination network in the search result category determination model, a first classification score for classifying the search text into each of the first search result categories according to the semantic features of the search text; According to the first classification score, a target search result category corresponding to the search text is determined in each of the first search result categories.

3. The method according to claim 2, characterized in that The step of determining, based on the semantic features of the search text, a first classification score for classifying the search text into each of the first search result categories includes: Determining a first similarity between a semantic feature of the search text and a semantic feature of each of the first search result categories; the semantic features of the first search result categories are stored in the first score determination network; A first classification score for classifying the search text into each of the first search result categories is determined according to the first similarity.

4. The method according to claim 2, characterized in that: Determining, in each of the first search result categories, a target search result category corresponding to the search text according to the first classification score includes: Normalizing the first classification score; A target search result category corresponding to the search text is determined in each of the first search result categories according to the normalized first category score and the first category score threshold.

5. The method according to claim 1, characterized in that Also includes: Obtaining a first identification text of the first search result; The length of the first identification text is less than a first length threshold; Determining a corresponding relationship between the first search result category and the first identification text according to a subordinate relationship between the first search result category and the first search result; training a first neural network according to the correspondence between the first identification text, the first search result category and the first identification text; The trained first neural network determines a model for the search result category.

6. The method according to claim 5, characterized in that The step of training the first neural network according to the correspondence between the first identification text, the first search result category and the first identification text includes: Extracting semantic features of the first identification text through a first feature extraction network in the first neural network; Determining, by a first score determination network in the first neural network, a second classification score for classifying the first identification text into each of the first search result categories according to the semantic features of the first identification text; The first feature extraction network and the parameters in the first score determination network are trained according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text.

7. The method according to claim 6, characterized in that The step of determining, according to the semantic feature of the first identification text, a second classification score for classifying the first identification text into each of the first search result categories includes: Determining a second similarity between the semantic feature of the first identification text and the semantic feature of each of the first search result categories; the semantic features of the first search result category are initialized in the first score determination network; the semantic features of the first search result category are adjusted during the model training process; Determine a second classification score for classifying the first identification text into each of the first search result categories according to the second similarity.

8. The method according to claim 6, characterized in that The step of training the first feature extraction network and the first score determination network parameters according to the second classification score until the approximate search result category corresponding to the first identification text represented by the second classification score satisfies the correspondence between the first search result category and the first identification text includes: Normalizing the second classification score; Determining, in each of the first search result categories, an approximate search result category corresponding to the first identification text according to the normalized second classification score and the second classification score threshold; The parameters in the first feature extraction network and the first score determination network are trained according to the approximate search result category corresponding to the first identification text until the approximate search result category corresponding to the first identification text satisfies the correspondence between the first search result category and the first identification text.

9. The method according to claim 1, characterized in that: Also includes: Acquire a second identification text of the first search result; the length of the second identification text is greater than a second length threshold; Extracting semantic features of the second identification text by using a semantic extraction model; The first search results are clustered according to the semantic features of the second identification text to obtain the subordinate relationship between the first search result category and the first search result.

10. The method according to claim 9, characterized in that Also includes: Obtaining sample search results and sample identification text of the sample search results; The length of the sample identification text is greater than a third length threshold; According to the sample identification text of the sample search results, a second neural network is trained; the trained second neural network is used to extract sample semantic features of the sample identification text and classify the sample identification text of the same sample search results into the same sample search results according to the sample semantic features; The semantic extraction model is generated according to the trained second neural network.

11. The method according to claim 10, characterized in that The obtaining of the sample search results and the sample identification text of the sample search results includes: Obtaining the sample search result and a third identification text of the sample search result; Rewrite the third identification text based on the semantic information of the third identification text to obtain a fourth identification text; The sample identification text is generated according to the third identification text and the fourth identification text.

12. The method according to claim 10, characterized in that The step of training a second neural network according to the sample identification text of the sample search results includes: Extracting sample semantic features of the sample identification text through a second feature extraction network in the second neural network; Determining, by a second score determination network in the second neural network, a third classification score for classifying the sample identification text into each of the sample search results according to the sample semantic features of the sample identification text; The second feature extraction network and the parameters in the second score determination network are trained according to the third classification score until the third classification score indicates that the sample identification text of the same sample search result is classified into the same sample search result.

13. The method according to claim 12, characterized in that The step of determining, according to the sample semantic features of the sample identification text, a third classification score for classifying the sample identification text into each of the sample search results includes: Determining a third similarity between the sample semantic feature of the sample identification text and the semantic feature of each of the sample search results; the semantic features of the sample search results are initialized in the second score determination network; the semantic features of the sample search results are adjusted during the model training process; A third classification score for classifying the sample identification text into each of the sample search results is determined according to the third similarity.

14. The method according to claim 12, characterized in that The step of training the second feature extraction network and the parameters in the second score determination network according to the third classification score until the third classification score indicates that the sample identification text of the same sample search result is classified into the same sample search result includes: Normalizing the third classification score; Determining, in each of the sample search results, an approximate search result corresponding to the sample identification text according to the normalized third classification score and the third classification score threshold; The parameters in the second feature extraction network and the second score determination network are trained according to the approximate search results corresponding to the sample identification text until the sample identification text of the same sample search result is classified into the same sample search result.

15. A search result determination device, characterized in that: include: A category determination unit, configured to obtain a user's search text, and determine a target search result category corresponding to the search text in each predetermined first search result category according to the semantic features of the search text through a search result category determination model; A first result determination unit, configured to determine the first search results subordinate to the target search result category according to the target search result category and a predetermined subordinate relationship between the first search result category and the first search result; wherein the semantic information of the first search results subordinate to the same first search result category is associated; The second result determination unit is configured to determine a target search result corresponding to the search text according to the first search result belonging to the target search result category.

16. An electronic device, characterized in that: include: processor; as well as, A memory configured to store computer executable instructions, which, when executed, cause the processor to implement the steps of the method according to any one of claims 1 to 14.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store computer-executable instructions, and the computer-executable instructions, when executed by a processor, implement the steps of the method described in any one of claims 1 to 14.

18. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 14 are implemented.

Citation Information

Patent Citations

  • Search statement processing method, device and server

    CN110516226A

  • Classification processing and data searching method and device, electronic equipment and intelligent loudspeaker box

    CN112241490A

  • Health information searching system and method

    CN114510619A

  • Search attribution method and device, computer equipment and storage medium

    CN115795094A