CN117851546A

CN117851546A - Resource retrieval method, training method, device, electronic equipment, storage medium and program product

Info

Publication number: CN117851546A
Application number: CN202410102934.XA
Authority: CN
Inventors: 冯朝兵; 林坤海; 汪洋; 黄飞; 贺登武; 孙琳; 史须勇; 邢越
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2024-01-24
Filing date: 2024-01-24
Publication date: 2024-04-09

Abstract

The disclosure provides a resource retrieval method, a training device, electronic equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, in particular to the technical field of intelligent retrieval, the technical field of big data and the technical field of deep learning. The specific implementation scheme is as follows: in response to receiving the search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and retrieval interaction operation; according to the semantic relevance characteristics and the interaction attribute characteristics, determining associated search information corresponding to the search information from a preset search information base; and searching the resources according to the associated search information to obtain the target resources.

Description

Resource retrieval method, training method, device, electronic equipment, storage medium and program product

Technical Field

The present disclosure relates to the field of artificial intelligence, and in particular, to the fields of intelligent retrieval, big data technology, and deep learning.

Background

With the rapid development of internet technology, users can rapidly browse resource information such as news through terminal equipment such as smart phones, search information such as texts can be input into the terminal equipment based on requirements, and the terminal equipment can perform resource retrieval based on the search information so as to provide resources matched with retrieval requirements of the users and improve information acquisition efficiency of the users.

Disclosure of Invention

The present disclosure provides a resource retrieval method, training method, apparatus, electronic device, storage medium, and program product.

According to an aspect of the present disclosure, there is provided a resource retrieval method, including: in response to receiving the search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and retrieval interaction operation; according to the semantic relevance characteristics and the interaction attribute characteristics, determining associated search information corresponding to the search information from a preset search information base; and searching the resources according to the associated search information to obtain the target resources.

According to another aspect of the present disclosure, there is provided a training method of a deep learning model, including: obtaining a training sample, wherein the training sample comprises sample search information, sample preset search information and a sample label, the sample label comprises a semantic correlation label and an interaction attribute label, the semantic correlation label represents semantic correlation between the sample search information and the sample preset search information, and the interaction attribute label represents correlation between the sample interaction search information, the sample preset search information and historical retrieval interaction operation; inputting sample search information and sample preset search information into an initial deep learning model, and outputting sample semantic relevance characteristics and sample interaction attribute characteristics; and training the initial deep learning model according to the sample semantic relevance features, the sample interaction attribute features, the semantic relevance labels and the interaction attribute labels to obtain a trained deep learning model.

According to another aspect of the present disclosure, there is provided a resource retrieval device including: the detection module is used for responding to the received search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and search interaction operation; the associated search information determining module is used for determining associated search information corresponding to the search information from a preset search information base according to the semantic relevance characteristics and the interaction attribute characteristics; and the resource retrieval module is used for retrieving the resources according to the associated search information to obtain target resources.

According to another aspect of the present disclosure, there is provided a training apparatus of a deep learning model, including: the training sample acquisition module is used for acquiring a training sample, the training sample comprises sample search information, sample preset search information and a sample label, the sample label comprises a semantic correlation label and an interactive attribute label, the semantic correlation label represents semantic correlation between the sample search information and the sample preset search information, and the interactive attribute label represents correlation between the sample interactive search information, the sample preset search information and historical retrieval interactive operation; the feature acquisition module is used for inputting sample search information and sample preset search information into an initial deep learning model and outputting sample semantic relevance features and sample interaction attribute features; and the training module is used for training the initial deep learning model according to the sample semantic relevance characteristics, the sample interaction attribute characteristics, the semantic relevance labels and the interaction attribute labels to obtain a trained deep learning model.

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

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

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements a method provided according to embodiments of the present disclosure.

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

Drawings

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

FIG. 1 schematically illustrates an exemplary system architecture to which resource retrieval methods and apparatus may be applied, according to embodiments of the present disclosure;

FIG. 2 schematically illustrates a flow chart of a resource retrieval method according to an embodiment of the disclosure

FIG. 3 schematically illustrates a schematic diagram of semantic relevance detection and interaction attribute detection of search information according to an embodiment of the present disclosure;

Fig. 4 schematically illustrates a schematic diagram of neighbor retrieval in a search information topology according to an embodiment of the present disclosure;

FIG. 5 schematically illustrates a schematic diagram of determining a plurality of target resources associated with a target resource index in accordance with an embodiment of the present disclosure;

FIG. 6 schematically illustrates a flow chart of a training method of a deep learning model according to an embodiment of the present disclosure;

FIG. 7 schematically illustrates a schematic diagram of a training method of a deep learning model according to an embodiment of the present disclosure;

FIG. 8 schematically illustrates a block diagram of a resource retrieval device according to an embodiment of the present disclosure;

FIG. 9 schematically illustrates a block diagram of a training apparatus of a deep learning model according to an embodiment of the present disclosure; and

fig. 10 schematically illustrates a block diagram of an electronic device adapted to implement a resource retrieval method, a training method, according to an embodiment of the disclosure.

Detailed Description

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

In the technical scheme of the disclosure, the acquisition, storage, application and the like of the related personal information of the user all conform to the regulations of related laws and regulations, necessary security measures are taken, and the public order harmony is not violated.

The user can search resources by inputting search information such as text in a terminal device such as a smart phone, and the terminal device can search resources based on the search information to provide resources matched with the search requirements of the user. Accordingly, the resource throwing parties such as advertisers can set the mapping relation between the resources to be thrown and preset search information (or bid search terms) in a bid mode, so that the recall rate of the thrown resources is improved, and the thrown resources can be accurately pushed to users who search for the resources based on the preset search information. The inventor finds that semantic difference may exist between the preset search information and the search information input by the user, so that the matching degree between the resource searched according to the search information and the preset search information or the search information may be low, and it is difficult to accurately meet the search requirement of the user.

Embodiments of the present disclosure provide a resource retrieval method, training method, apparatus, electronic device, storage medium, and program product. The resource retrieval method comprises the following steps: in response to receiving the search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and retrieval interaction operation; according to the semantic relevance characteristics and the interaction attribute characteristics, determining associated search information corresponding to the search information from a preset search information base; and searching the resources according to the associated search information to obtain the target resources.

According to the embodiment of the disclosure, by carrying out semantic relevance detection on the search information, the obtained semantic relevance features can represent semantic relevance between the search information input by the target object and preset search information, and the interactive attribute features obtained by carrying out interactive attribute detection on the search information can represent the relevance degree of the resource recalled by the search information for executing the search interactive operation by the target object, so that the associated search information is determined from the preset search information base according to the semantic relevance features and the interactive attribute features, the semantic relevance degree between the associated search information and the search information can be improved, and the associated search information can be matched with the search interactive operation attribute of the target object, so that resource search is carried out according to the associated search information, the obtained target resource can more accurately meet the search requirement of the target object, and meanwhile, the probability of the target object for executing the operation on the target resource is improved, and the accuracy and the search efficiency of the resource search are improved.

Fig. 1 schematically illustrates an exemplary system architecture to which resource retrieval methods and apparatuses may be applied, according to embodiments of the present disclosure.

It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios. For example, in another embodiment, an exemplary system architecture to which the resource searching method and apparatus may be applied may include a terminal device, but the terminal device may implement the resource searching method and apparatus provided by the embodiments of the present disclosure without interacting with a server.

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

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as a knowledge reading class application, a web browser application, a search class application, an instant messaging tool, a mailbox client and/or social platform software, etc. (as examples only).

The terminal devices 101, 102, 103 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.

The server 105 may be a server providing various services, such as a background management server (by way of example only) providing support for content browsed by the user using the terminal devices 101, 102, 103. The background management server may analyze and process the received data such as the user request, and feed back the processing result (e.g., the web page, information, or data obtained or generated according to the user request) to the terminal device.

It should be noted that, the resource searching method provided by the embodiments of the present disclosure may be generally performed by the terminal device 101, 102, or 103. Accordingly, the resource retrieving apparatus provided by the embodiments of the present disclosure may also be provided in the terminal device 101, 102, or 103.

Alternatively, the resource retrieval method provided by the embodiments of the present disclosure may also be generally performed by the server 105. Accordingly, the resource retrieval device provided by the embodiments of the present disclosure may be generally provided in the server 105. The resource retrieval method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the resource retrieving apparatus provided by the embodiments of the present disclosure may also be provided in a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.

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

Fig. 2 schematically illustrates a flow chart of a resource retrieval method according to an embodiment of the present disclosure.

As shown in fig. 2, the resource retrieval method includes operations S210 to S230.

In response to receiving the search information, semantic relevance detection and interaction attribute detection are performed on the search information, resulting in semantic relevance features and interaction attribute features in operation S210.

In operation S220, associated search information corresponding to the search information is determined from a preset search information base according to the semantic relevance feature and the interaction attribute feature.

In operation S230, resource retrieval is performed according to the associated search information to obtain a target resource.

According to an embodiment of the present disclosure, the search information may include any type of information such as text characters, text words, characters, etc. generated by the target object based on the input operation, and the embodiment of the present disclosure does not limit the specific type of the search information.

According to the embodiment of the present disclosure, the resources may include any type of resources that can be browsed based on a terminal device such as a smart phone, including news resources, advertisement resources, video resources, etc., and the embodiment of the present disclosure does not limit the specific type of the resources, and a person skilled in the art may select according to actual needs.

According to embodiments of the present disclosure, semantic relevance features may characterize semantic relevance between search information and preset search information in a preset search information library, and interaction attribute features may characterize relevance between search information and retrieval interactions.

It should be noted that, the search interactive operation may include any type of operation, such as a clicking operation, a box selection operation, etc., and the embodiment of the present disclosure does not limit a specific operation type of the search interactive operation, and those skilled in the art may select according to actual needs.

According to embodiments of the present disclosure, the interaction attribute feature may characterize a correlation between search information, preset search information, and a retrieve interaction, for example, may characterize a probability attribute of a retrieve interaction being performed by a target object based on resources retrieved by the preset search information associated with the search information.

According to an embodiment of the disclosure, performing semantic relevance detection on the search information may include performing semantic feature extraction on the search information based on a pre-trained neural network model, resulting in semantic relevance features.

In one embodiment of the present disclosure, the pre-trained neural network model may be pre-trained based on the sample search information and the preset search information having an interactive relationship with the sample search information, so that the pre-trained neural network model may perform semantic feature extraction on the search information, and the obtained semantic relevance features may be capable of characterizing the semantic relevance degree between the search information and the preset search information.

According to embodiments of the present disclosure, interactive attribute detection of search information may include processing the search information based on a pre-trained attention network model. For example, the attention network model may be pre-trained based on the sample search information and the preset search information having an interactive relationship with the sample search information, so that the pre-trained attention network can learn the association attribute between the sample search information and the preset search information and the retrieval interactive operation, and thus the search information is processed by using the pre-trained attention network, and the obtained interactive attribute features can characterize the association attribute between the search information and the preset search information and the retrieval interactive operation. In this way, the associated search information is determined according to the semantic relevance characteristics and the interaction attribute characteristics, so that the obtained associated search information can accurately represent the semantic attribute of the search information and the associated attribute between the search information and the associated search information and the search interaction operation, the target resource obtained through search according to the associated search information can meet the search requirement of the target object and the resource release requirement of the resource release party, the resource search accuracy and the resource release efficiency are further improved, and the user experience is improved.

According to an embodiment of the present disclosure, performing semantic relevance detection and interaction attribute detection on search information, obtaining semantic relevance features and interaction attribute features may include: extracting semantic features of the search information to obtain search semantic features; extracting semantic correlation features from the searched semantic features to obtain semantic correlation features; and extracting the interaction attribute features from the search semantic features to obtain the interaction attribute features.

According to the embodiment of the disclosure, semantic feature extraction is performed on the search information, and the search information can be processed based on a pre-trained encoder to obtain search semantic features. The encoder may be constructed based on an encoder of an attention network algorithm (e.g., a transducer algorithm).

According to embodiments of the present disclosure, extracting semantic relevance features from search semantic features may include inputting the search semantic features into a pre-trained semantic relevance detection layer, outputting the semantic relevance features. The semantic relevance detection layer can be constructed based on a neural network algorithm, for example, the semantic relevance detection layer can be constructed based on a convolutional neural network algorithm, an attention network algorithm and the like, the specific algorithm type for constructing the semantic relevance detection layer is not limited in the embodiment of the disclosure, and a person skilled in the art can select according to actual requirements.

According to embodiments of the present disclosure, performing interactive attribute feature extraction on search semantic features may include inputting the search semantic features into a pre-trained interactive attribute detection layer, outputting interactive attribute features. The interactive attribute detection layer can be constructed based on a neural network algorithm, for example, can be constructed based on a convolutional neural network algorithm, an attention network algorithm and the like, the specific algorithm type for constructing the interactive attribute detection layer is not limited in the embodiment of the disclosure, and a person skilled in the art can select according to actual requirements.

Fig. 3 schematically illustrates a schematic diagram of semantic relevance detection and interaction attribute detection of search information according to an embodiment of the present disclosure.

As shown in fig. 3, the deep learning model 300 may include a semantic feature extraction layer 310, a semantic relevance detection layer 320, an interaction attribute detection layer 330, and a fusion layer 340. The search information 301 may be input to the semantic feature extraction layer 310, outputting search semantic features. The semantic feature extraction layer 310 may be constructed based on an attention network algorithm, for example, the semantic feature extraction layer 310 may be constructed based on an Ernie algorithm.

As shown in fig. 3, the search semantic features may be input to the semantic relevance detection layer 320, outputting semantic relevance features 321. The search semantic features may also be input to the interaction attribute detection layer 330, outputting interaction attribute features 331. The semantic relevance detection layer 320 and the interaction attribute detection layer 330 may be built based on pre-trained deep neural network (Deep Neural Networks, DNN) models. The semantic relevance features 321 and interaction attribute features 331 are input into the fusion layer 340 and the search fusion features 302 can be output. The fusion layer 340 may be constructed based on a neural network algorithm, for example, the fusion layer 340 may be constructed based on a multi-layer perceptron (MLP, multilayer Perceptron) algorithm, or the fusion layer 340 may also be constructed based on other types of algorithms, for example, the fusion layer 340 may be constructed based on an additive function. The embodiment of the disclosure does not limit the specific manner of constructing the fusion layer, and may be selected based on actual requirements.

According to an embodiment of the present disclosure, a search information topology is constructed using the following manner: carrying out semantic relevance detection and interaction attribute detection on preset search information to obtain preset semantic relevance characteristics and preset interaction attribute characteristics; performing feature fusion on the preset semantic relativity features and the preset interaction attribute features to obtain preset search fusion features; determining a preset search information node according to preset search fusion characteristics; and constructing a search information topological graph according to the preset search information nodes.

According to the embodiment of the disclosure, the semantic relevance detection and the interaction attribute detection can be performed on the preset search information based on the pre-trained semantic feature extraction layer, the semantic relevance detection layer and the interaction attribute detection layer, and the preset semantic relevance feature and the preset interaction attribute feature are subjected to feature fusion based on the feature fusion mode of the semantic relevance feature and the interaction attribute feature, so that the preset search fusion feature is obtained, the preset search information can be processed based on the same mode of processing the search information, and a preset search fusion feature set is constructed according to the obtained preset search fusion feature. Therefore, similarity calculation can be carried out on the search fusion feature and the preset search fusion feature, the preset search fusion feature matched with the search fusion feature is used as the associated search fusion feature, and associated search information is further determined.

According to the embodiment of the disclosure, the preset search information node is determined according to the preset search fusion characteristics, so that the preset search information node can represent semantic relativity between the preset search information and other preset search information or currently received search information, and can also represent retrieval operation attribute relationship between the preset search information and other preset search information or currently received search information, and therefore, the attribute of multiple dimensions of the preset search information can be intuitively represented based on the preset search information node. Therefore, the search information topological graph is constructed according to the preset search information nodes, the similarity degree among the preset search information can be intuitively represented based on the preset search information nodes in the search information topological graph on the basis of integrating the attributes of the multiple dimensions, and the follow-up search information similar to the search information can be conveniently queried based on the search fusion characteristics.

According to an embodiment of the present disclosure, the preset search information may include a plurality of search information, and the above manner of constructing the search information topology map may further include: performing correlation detection on the plurality of preset search information to obtain preset correlation detection results among the plurality of preset search information; and determining the side relationship among a plurality of preset search information nodes according to the preset correlation detection result.

According to an embodiment of the present disclosure, constructing a search information topology map according to a preset search information node includes: and constructing a search information topological graph according to the preset search information nodes and the preset side relationship.

According to an embodiment of the disclosure, performing semantic relevance detection on a plurality of preset search information may include processing at least two of the plurality of preset search information based on a pre-trained semantic classification model, and the obtained preset relevance detection result may represent whether semantics between different preset search information are the same. According to the preset correlation detection result, the side relation among a plurality of preset search information nodes is determined, and further according to the search information topological graph constructed by the preset search information nodes and the side relation, different preset search information with the same semantics or higher semantic similarity can be intuitively represented.

According to an embodiment of the present disclosure, determining associated search information corresponding to the search information from a preset search information base according to the semantic relevance feature and the interaction attribute feature may include: feature fusion is carried out on semantic relevance features and interaction attribute features, and search fusion features are obtained; determining a search information node according to the search fusion characteristics; performing neighbor search in a search information topological graph according to the search information nodes to obtain associated search information nodes, wherein the search information topological graph comprises preset search nodes corresponding to preset search information and side relations among a plurality of preset search nodes, and the side relations represent semantic relativity among a plurality of preset search information; and determining the associated search information according to the associated search information node.

According to embodiments of the present disclosure, determining a search information node from a search fusion feature may include converting the search fusion feature to a search information node in a search information topology map. Or the method also comprises the steps of detecting the similarity between the search fusion feature and the preset search fusion feature to obtain a similar search fusion feature with higher similarity to the search fusion feature, and determining the preset search node corresponding to the similar search fusion feature as the search information node.

According to the embodiment of the disclosure, the search scope can be reduced by performing neighbor search in the search information topological graph according to the search information nodes, so that the massive preset search information nodes in the search information topological graph are prevented from being subjected to full-scale search, and the search efficiency of the related search information nodes is improved. Because the preset search information node is determined based on the preset search fusion characteristics corresponding to the preset search information, the searched associated search information can be related to the search information in the semantic relevance, the search operation attribute relationship and other attributes of multiple dimensions, and further the resource searching is performed based on the associated search information, so that the obtained target resource can relatively accurately meet the search requirement of the target object, and meanwhile, the resource throwing requirement of a resource throwing party can be met, and the search efficiency and the user experience are improved.

According to an embodiment of the present disclosure, performing neighbor search in a search information topology according to a search information node, obtaining an associated search information node may include: according to the search information nodes, performing neighbor search in the search information topological graph to obtain a plurality of candidate associated search information nodes; carrying out semantic similarity detection on the search information and candidate associated search information associated with the candidate associated search information node to obtain a similarity detection result; and determining associated search information from the plurality of candidate associated search information according to the similarity detection result.

According to the embodiments of the present disclosure, candidate associated search information corresponding to a candidate associated search information node may have a high semantic similarity with search information, but in the case where a semantic correlation difference between a plurality of candidate associated search information is large, it is difficult for the candidate associated search information to accurately match with the semantics of the search information. Therefore, the semantic similarity detection is carried out on the search information and the candidate association search information, and the obtained similarity detection result can further accurately represent the semantic correlation between the search information and the candidate association search information. And determining the associated search information from the plurality of candidate associated search information according to the similarity detection result, so that the semantic similarity degree between the associated search information and the search information can be improved, resource retrieval can be performed based on the associated search information serving as the same semantic information of the search information, and the accuracy of resource retrieval is improved.

Fig. 4 schematically illustrates a schematic diagram of neighbor retrieval in a search information topology according to an embodiment of the present disclosure.

As shown in fig. 4, the search information topology 400 may include a plurality of preset search information nodes and a side relationship between the plurality of search information nodes. In case of determining the search information node 401, a first cluster 410 may be obtained by performing neighbor search according to the search information node 401. The first cluster 410 may include a plurality of level 1 candidate search information nodes associated with the search information node 401. A plurality of level 1 candidate search information nodes associated with search information node 401 may be determined, for example, based on the side relationships characterized by the solid and dashed lines.

As shown in fig. 4, the pre-trained semantic classification model may be used to process the search information corresponding to the search information node 401, and the level 1 candidate associated search information corresponding to each of the plurality of level 1 candidate search information nodes in the first cluster 410, so as to perform semantic similarity detection, and determine, according to the similarity detection result, a second cluster 420 containing the level 2 candidate associated search information node from the first cluster 410.

As shown in fig. 4, the pre-trained semantic classification model may also be used to iteratively process the search information corresponding to the search information node 401, and the level 2 candidate associated search information corresponding to each of the plurality of level 2 candidate search information nodes in the second cluster 420, so as to perform semantic similarity detection, and determine a third cluster 430 including the level 3 candidate associated search information node from the second cluster 420 according to the similarity detection result. A fourth cluster 440 is also correspondingly available. The level 3 candidate associated search information node in the third cluster 430 may be regarded as an associated search information node, and associated search information may be determined according to the associated search information node.

According to the embodiment of the disclosure, the resource retrieval method constructs the search information topological graph according to the semantic relation among the preset search information and the preset search fusion characteristic, performs preliminary clustering based on neighbor retrieval, and can reduce the calculation overhead and the calculation time consumption generated by the full-volume retrieval compared with full-space enumeration of N complexity. The candidate search information obtained by screening is further subjected to semantic judgment and verification based on semantic similarity detection, so that the search requirement that the topological connected graph does not meet the same semantic is reduced due to synonymous error transfer, and therefore the semantic similarity degree between the associated preset search information and the search information in the same cluster can be improved based on iterative synonymous judgment and verification, the storage space is saved, and the search efficiency is improved.

According to an embodiment of the present disclosure, performing resource retrieval according to the associated search information, obtaining the target resource may include: determining a target resource index according to the associated search information; and determining a target resource associated with the target resource index.

According to the embodiment of the disclosure, part or all of the associated search information can be used as the target resource index, so that the target resource is queried according to the target resource index. Or a resource index field may also be set for preset search information, and the target resource may be determined through the resource index field corresponding to the associated search information. The embodiment of the present disclosure does not limit the manner of determining the target resource index, and those skilled in the art may select according to actual requirements.

According to an embodiment of the present disclosure, the associated search information includes a plurality of associated search information associated with the same target resource index.

According to an embodiment of the present disclosure, determining a target resource associated with a target resource index includes: a plurality of target resources associated with the target resource index is determined.

According to the embodiment of the disclosure, the storage space for storing the resource index data can be reduced by establishing the mapping relation between the same target resource index and the plurality of associated search information, and meanwhile, the plurality of target resources are determined by the same target resource index, so that the retrieval efficiency of retrieving the target resources can be improved, and the retrieval timeliness is improved.

Fig. 5 schematically illustrates a schematic diagram of determining a plurality of target resources associated with a target resource index according to an embodiment of the present disclosure.

As shown in figure 5 of the drawings, the plurality of associated preset search information corresponding to the search information may include 'fresh flower express' 511 512 percent of fresh flower express delivery the term "express flowers" 51N. Each of the "flower express" 511, the "flower express" 512, the "express flowers" 51N may be associated with a target resource, for example, the "flower express" 511 may be related to target resources such as the 1 st target resource 521 and the 2 nd target resource 522 based on the resource index. The search information preset by associating a plurality of links can comprise 'fresh flower express' 511, 'fresh flower express' 512 the resource index of each to "express flower" 51N is compressed to the same target resource index, can make 'fresh flower express' 511, 'fresh flower express' 512 the third party is to "express flowers" associated with the same target resource index. The 1 st target resource 521, the 2 nd target resource 522, and the..once..th to nth target resources 52N can be obtained by searching for resources based on the target resource index. Therefore, the resource indexes of the plurality of associated search information can be compressed, so that the storage space of the resource indexes is saved, and the resource retrieval efficiency is improved.

According to an embodiment of the present disclosure, the target resource corresponds to the associated search information, and the plurality of target resources are arranged based on recall order attributes corresponding to each of the plurality of associated search information.

According to an embodiment of the present disclosure, the resource retrieval method may further include: and displaying the plurality of target resources based on the arrangement sequence attribute corresponding to each of the plurality of target resources.

According to embodiments of the present disclosure, the recall order attributes may include resource placement settings according to actual requirements. Alternatively, the recall sequence attribute may be determined based on the respective importance levels of the associated search information, and the embodiment of the disclosure does not limit the manner of determining the recall sequence attribute corresponding to the associated search information, and those skilled in the art may select the recall sequence attribute according to actual requirements.

According to the embodiment of the disclosure, based on the arrangement sequence attribute corresponding to each of the plurality of target resources, displaying the plurality of target resources may include sorting the plurality of target resources according to a list form, so that the target object may preview the target resources according to the sorting results of the plurality of target resources, or click on the target resources of interest to view the resources.

Fig. 6 schematically illustrates a flowchart of a training method of a deep learning model according to an embodiment of the present disclosure.

As shown in fig. 6, the training method of the deep learning model includes operations S610 to S630.

In operation S610, a training sample is obtained, the training sample includes sample search information, sample preset search information, and a sample tag, the sample tag includes a semantic correlation tag and an interaction attribute tag, the semantic correlation tag characterizes semantic correlation between the sample search information and the sample preset search information, and the interaction attribute tag characterizes correlation between the sample interaction search information, the sample preset search information, and the history retrieval interaction operation.

In operation S620, sample search information and sample preset search information are input into the initial deep learning model, and sample semantic relevance features and sample interaction attribute features are output.

In operation S630, an initial deep learning model is trained according to the sample semantic relevance features, the sample interaction attribute features, the semantic relevance labels, and the interaction attribute labels, and a trained deep learning model is obtained.

According to embodiments of the present disclosure, the sample search information may include search information input by the target object in the history period. The sample preset search information may include preset search terms (or bid) set by the resource dispenser through bidding or the like, or the sample preset search information may also include search terms determined through other manners, which are not limited by embodiments of the present disclosure.

According to the embodiment of the present disclosure, the initial deep learning model may be constructed based on a deep learning algorithm, for example, the initial deep learning model may be constructed based on an attention network algorithm, but not limited thereto, and the initial deep learning model may be constructed based on other types of deep learning algorithms, which is not limited thereto.

According to an embodiment of the present disclosure, the semantic relevance tag may characterize semantic relevance between the sample search information and the sample preset search information, for example, may characterize the sample search information and the sample preset search information to have the same semantic based on the semantic relevance tag "1", and may characterize the sample search information and the sample preset search information to have different semantic based on the semantic relevance tag "0".

According to embodiments of the present disclosure, the interaction attribute tags may characterize interaction-operation correlations (or retrieval-operation attributes) between sample search information, sample preset search information, and retrieval interactions in a historical time period. For example, the interaction attribute tag "1" may characterize searching based on sample search information in a historical period, and perform a search interaction operation such as a click operation by a target object according to a resource recalled by sample preset search information. For another example, the interaction attribute tag "0" may characterize that the search is performed based on the sample search information in the history period, and the resources recalled according to the sample preset search information are not performed by the target object for the search interaction. Alternatively, the interaction attribute tag "0" may characterize that the search information is retrieved based on a sample during a historical period of time, and the search information recall resource is not preset based on the sample.

According to the embodiment of the disclosure, the sample search information and the sample preset search information are input into the initial deep learning model, the sample semantic relevance features and the sample interaction attribute features are output, so that the initial deep learning model can detect semantic attributes under the condition that the sample search information and the sample preset search information are fully fused, the initial deep learning model is trained by utilizing the obtained sample semantic relevance features and semantic relevance labels, and the trained deep learning model can fully learn the semantic relevance between the sample search information and the sample preset search information. The initial deep learning model is trained by using the sample interaction attribute characteristics and the interaction attribute labels, so that the deep learning model obtained through training can fully learn the search interaction operation attribute relationship between the sample search information and the sample preset search information. Therefore, the trained deep learning model can accurately detect semantic relevance and interaction attribute of the search information, the matching degree between the search information and the associated search information is improved, resource retrieval is carried out according to the associated search information, the matching degree between the retrieved target resource and the search information can be improved, recall level of the target resource is improved, and retrieval efficiency is improved.

According to embodiments of the present disclosure, the trained deep learning model may be applied to the resource retrieval method provided according to embodiments of the present disclosure. For example, the received search information may be input into a trained deep learning model, outputting semantic relevance features and interaction attribute features.

It should be noted that, the preset search information in the preset search information base may include at least one item of sample search information and sample preset search information. The deep learning model obtained after training can also be used for processing preset search information to obtain preset search fusion characteristics.

According to the embodiment of the disclosure, the semantic relevance labels and the interaction attribute labels between the sample search information and the sample preset search information can be determined based on a manual labeling mode. But is not limited thereto, semantic relevance labels and interaction attribute labels may be determined in other ways, as embodiments of the present disclosure are not limited thereto.

According to an embodiment of the present disclosure, the training method of the deep learning model may further include: processing sample search information and sample preset search information based on a pre-trained semantic classification model to obtain semantic classification results between the sample search information and the sample preset search information; and determining the semantic relevance labels according to the semantic classification results.

According to the embodiment of the disclosure, the semantic classification model may be constructed based on a deep learning algorithm, for example, may be constructed based on a convolutional neural network algorithm, but is not limited thereto, and may be constructed based on other types of deep learning algorithms, which is not limited thereto.

According to embodiments of the present disclosure, the semantic classification result may characterize whether semantic attributes between sample search information and sample preset search information are the same. Sample search information and sample preset search information are processed based on a pre-trained semantic classification model, a semantic relevance label is determined according to an obtained semantic classification result, and an initial deep learning model can be trained by utilizing the semantic classification model based on a knowledge distillation principle so as to improve the detection precision of the trained deep learning model.

According to an embodiment of the present disclosure, an initial deep learning model includes a semantic feature extraction layer, a semantic relevance detection layer, and an interaction attribute detection layer.

According to embodiments of the present disclosure, the semantic feature extraction layer may be constructed based on an attention network algorithm, for example, the semantic feature extraction layer may be constructed based on an Ernie algorithm. The semantic relevance detection layer and the interaction attribute detection layer may be constructed based on a pre-trained neural network model, for example, may be constructed based on a deep neural network (Deep Neural Networks, DNN) model.

According to an embodiment of the present disclosure, inputting sample search information and sample preset search information into an initial deep learning model, outputting sample semantic relevance features and sample interaction attribute features may include: inputting sample search information and sample preset search information into a semantic feature extraction layer, and outputting sample search semantic features and sample preset search semantic features; inputting the sample search semantic features and the sample preset search semantic features into a semantic relevance detection layer, and outputting sample semantic relevance features; and inputting the sample search semantic features and the sample preset search semantic features into an interaction attribute detection layer, and outputting sample interaction attribute features.

According to embodiments of the present disclosure, training an initial deep learning model from sample semantic relevance features, sample interaction attribute features, semantic relevance labels, and interaction attribute labels may include: determining a first loss value according to the sample semantic correlation characteristics and the semantic correlation labels; determining a second loss value according to the sample interaction attribute characteristics and the interaction attribute labels; and training an initial deep learning model based on the first loss value and the second loss value.

Fig. 7 schematically illustrates a schematic diagram of a training method of a deep learning model according to an embodiment of the present disclosure.

As shown in fig. 7, the initial deep learning model 700 may include a semantic feature extraction layer 710, a semantic relevance detection layer 720, and an interaction attribute detection layer 730. The sample search information 701 and the sample preset search information 702 may be input to the semantic feature extraction layer 710, respectively, outputting a sample search semantic feature 711 and a sample preset search semantic feature 712. The semantic feature extraction layer 710 may be constructed based on an attention network algorithm, for example, the semantic feature extraction layer 710 may be constructed based on an Ernie algorithm.

As shown in fig. 7, sample search semantic features 711 and sample preset search semantic features 712 may be input to the semantic relevance detection layer 720 in order to enable the initial deep learning model 700 to fully learn the semantic relevance between the sample search information 701 and the sample preset search information 702, outputting sample semantic relevance features 721. Sample search semantic features 711 and sample preset search semantic features 712 may be input to the interaction attribute detection layer 730 in order to enable the initial deep learning model 700 to fully learn the interaction correlation between the sample search information 701, the sample preset search information 702, and the retrieve interaction, outputting sample interaction attribute features 731. The semantic relevance detection layer 720 and the interaction attribute detection layer 730 may be built based on pre-trained deep neural network (Deep Neural Networks, DNN) models. The sample semantic relevance features 721 and semantic relevance labels 703 may be processed according to a penalty function to obtain a first penalty value. And processing the sample interaction attribute feature 731 and the interaction attribute label 704 according to the loss function to obtain a second loss value, and determining a joint loss value according to the first loss value and the second loss value so as to adjust model parameters of the initial deep learning model 700 according to the joint loss value until the joint loss value converges to obtain a trained deep learning model.

According to the embodiment of the disclosure, according to the semantic relevance detection layer and the interaction attribute detection layer, semantic relevance targets and search operation relevance targets are respectively fitted, so that the deep learning model can accurately detect semantic relevance and interaction attributes of search information, and can detect the search information from multiple dimensions such as semantic relevance, search operation attribute relations and the like, the search fusion characteristics can measure the attributes of the search information from the multiple dimensions, the search information can be matched with preset search information in the multiple attribute dimensions, resource search is performed according to the matched associated search information, the search requirement of a target object and the resource throwing requirement of a resource throwing party can be further met, and the search efficiency is improved.

Fig. 8 schematically illustrates a block diagram of a resource retrieval device according to an embodiment of the present disclosure.

As shown in fig. 8, the resource retrieval device 800 includes: a detection module 810, an associated search information determination module 820, and a resource retrieval module 830.

The detection module 810 is configured to perform semantic relevance detection and interaction attribute detection on the search information in response to receiving the search information, to obtain semantic relevance features and interaction attribute features, where the semantic relevance features characterize semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features characterize relevance between the search information and search interaction operation.

The associated search information determining module 820 is configured to determine associated search information corresponding to the search information from a preset search information base according to the semantic relevance feature and the interaction attribute feature.

The resource retrieval module 830 is configured to perform resource retrieval according to the associated search information to obtain a target resource.

According to an embodiment of the present disclosure, the association search information determination module includes: the system comprises a search fusion feature obtaining sub-module, a search information node determining sub-module, an associated search information node obtaining sub-module and an associated search information determining sub-module.

And the search fusion feature obtaining sub-module is used for carrying out feature fusion on the semantic relevance features and the interaction attribute features to obtain search fusion features.

And the searching information node determining sub-module is used for determining searching information nodes according to the searching fusion characteristics.

The related search information node obtaining sub-module is used for carrying out neighbor search in a search information topological graph according to the search information nodes to obtain related search information nodes, wherein the search information topological graph comprises preset search nodes corresponding to preset search information and side relations among the preset search nodes, and the side relations represent semantic relativity among the preset search information.

And the associated search information determining sub-module is used for determining associated search information according to the associated search information nodes.

According to an embodiment of the present disclosure, the association search information node obtaining submodule includes: the candidate associated search information node obtaining unit, the similarity detection result obtaining unit, and the associated search information determining unit.

And the candidate associated search information node obtaining unit is used for carrying out neighbor search in the search information topological graph according to the search information nodes to obtain a plurality of candidate associated search information nodes.

The similarity detection result obtaining unit is used for carrying out semantic similarity detection on the search information and candidate associated search information associated with the candidate associated search information node to obtain a similarity detection result.

And an associated search information determining unit configured to determine associated search information from the plurality of candidate associated search information based on the similarity detection result.

According to an embodiment of the present disclosure, the preset search information includes a plurality of; the method further comprises the following steps: carrying out semantic relevance detection on a plurality of preset search information to obtain preset relevance detection results among the plurality of preset search information; determining the side relation among a plurality of preset search information nodes according to the preset correlation detection result; wherein, according to the preset search information node, constructing the search information topological graph comprises: and constructing a search information topological graph according to the preset search information nodes and the preset side relationship.

According to an embodiment of the present disclosure, a detection module includes: the system comprises a searching semantic feature obtaining sub-module, a semantic correlation feature obtaining sub-module and an interaction attribute feature obtaining sub-module.

And the searching semantic feature obtaining sub-module is used for extracting semantic features of the searching information to obtain searching semantic features.

The semantic correlation feature obtaining sub-module is used for extracting semantic correlation features of the search semantic features to obtain the semantic correlation features.

And the interaction attribute feature obtaining sub-module is used for extracting interaction attribute features of the search semantic features to obtain the interaction attribute features.

According to an embodiment of the present disclosure, a resource retrieval module includes: a target resource index determination sub-module and a target resource determination sub-module.

And the target resource index determining sub-module is used for determining the target resource index according to the associated search information.

And the target resource determining sub-module is used for determining target resources associated with the target resource index.

According to an embodiment of the present disclosure, the target resource determination submodule includes a target resource determination unit.

And the target resource determining unit is used for determining a plurality of target resources associated with the target resource index.

The resource retrieval device also comprises a display module.

And the display module is used for displaying the plurality of target resources based on the arrangement sequence attributes corresponding to the plurality of target resources.

Fig. 9 schematically illustrates a block diagram of a training apparatus of a deep learning model according to an embodiment of the present disclosure.

As shown in fig. 9, the training apparatus 900 for deep learning model includes: a training sample acquisition module 910, a feature acquisition module 920, and a training module 930.

The training sample obtaining module 910 is configured to obtain a training sample, where the training sample includes sample search information, sample preset search information, and a sample tag, and the sample tag includes a semantic correlation tag and an interaction attribute tag, where the semantic correlation tag characterizes semantic correlation between the sample search information and the sample preset search information, and the interaction attribute tag characterizes correlation between the sample interaction search information, the sample preset search information, and a history retrieval interaction operation.

The feature obtaining module 920 is configured to input the sample search information and the sample preset search information into an initial deep learning model, and output a sample semantic relevance feature and a sample interaction attribute feature.

The training module 930 is configured to train the initial deep learning model according to the sample semantic relevance feature, the sample interaction attribute feature, the semantic relevance tag, and the interaction attribute tag, and obtain a trained deep learning model.

According to an embodiment of the present disclosure, the feature obtaining module includes: the system comprises a first feature obtaining sub-module, a sample semantic relevance feature obtaining sub-module and a sample interaction attribute feature obtaining sub-module.

The first feature obtaining sub-module is used for inputting the sample search information and the sample preset search information into the semantic feature extraction layer and outputting the sample search semantic features and the sample preset search semantic features.

The sample semantic relevance feature obtaining sub-module is used for inputting the sample search semantic features and the sample preset search semantic features into the semantic relevance detection layer and outputting the sample semantic relevance features.

The sample interaction attribute feature obtaining sub-module is used for inputting the sample search semantic features and the sample preset search semantic features into the interaction attribute detection layer and outputting the sample interaction attribute features.

According to an embodiment of the present disclosure, the training device further comprises: the semantic classification result obtaining module and the semantic relativity label determining module.

The semantic classification result obtaining module is used for processing the sample search information and the sample preset search information based on the pre-trained semantic classification model to obtain a semantic classification result between the sample search information and the sample preset search information.

The semantic relativity label determining module is used for determining the semantic relativity label according to the semantic classification result.

According to an embodiment of the present disclosure, a training module includes: the first loss value determination sub-module, the second loss value determination sub-module, and the training sub-module.

The first loss value determining submodule is used for determining a first loss value according to the sample semantic correlation characteristics and the semantic correlation labels;

the second loss value determining submodule is used for determining a second loss value according to the sample interaction attribute characteristics and the interaction attribute labels; and

and the training sub-module is used for training the initial deep learning model according to the first loss value and the second loss value.

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

According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executable by the at least one processor to enable the at least one processor to perform the method as described above.

According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as described above.

According to an embodiment of the present disclosure, a computer program product comprising a computer program which, when executed by a processor, implements a method as described above.

Fig. 10 schematically illustrates a block diagram of an electronic device adapted to implement a resource retrieval method, a training method, according to an embodiment of the disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

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

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

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

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

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

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

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

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

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

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

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

Claims

1. A resource retrieval method, comprising:

in response to receiving search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and search interaction operation;

Determining associated search information corresponding to the search information from the preset search information base according to the semantic relevance characteristics and the interaction attribute characteristics; and

and searching the resources according to the associated search information to obtain target resources.

2. The method of claim 1, wherein the determining associated search information corresponding to the search information from the preset search information base according to the semantic relevance feature and the interaction attribute feature comprises:

feature fusion is carried out on the semantic relevance features and the interaction attribute features, and search fusion features are obtained;

determining a search information node according to the search fusion characteristics;

performing neighbor search in a search information topological graph according to the search information nodes to obtain associated search information nodes, wherein the search information topological graph comprises preset search nodes corresponding to preset search information and side relations among a plurality of preset search nodes, and the side relations represent semantic relativity among a plurality of preset search information; and

and determining the associated search information according to the associated search information node.

3. The method of claim 2, wherein the performing neighbor search in a search information topology according to the search information node, to obtain an associated search information node comprises:

performing neighbor search in a search information topological graph according to the search information nodes to obtain a plurality of candidate associated search information nodes;

carrying out semantic similarity detection on the search information and candidate associated search information associated with the candidate associated search information node to obtain a similarity detection result; and

and determining the associated search information from a plurality of candidate associated search information according to the similarity detection result.

4. The method of claim 2, wherein the search information topology map is constructed using:

carrying out semantic relevance detection and interaction attribute detection on the preset search information to obtain preset semantic relevance characteristics and preset interaction attribute characteristics;

performing feature fusion on the preset semantic relativity feature and the preset interaction attribute feature to obtain a preset search fusion feature;

determining the preset search information node according to the preset search fusion characteristics; and

And constructing the search information topological graph according to the preset search information nodes.

5. The method of claim 4, wherein the preset search information comprises a plurality of;

the method further comprises the following steps:

carrying out semantic relevance detection on a plurality of preset search information to obtain preset relevance detection results among the plurality of preset search information; and

determining the side relation among a plurality of preset searching information nodes according to the preset correlation detection result;

wherein, the constructing the search information topological graph according to the preset search information node includes:

and constructing the search information topological graph according to the preset search information nodes and the side relation.

6. The method of claim 1, wherein the performing semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features comprises:

extracting semantic features of the search information to obtain search semantic features;

extracting semantic relevance features from the search semantic features to obtain the semantic relevance features; and

and extracting the interaction attribute features from the search semantic features to obtain the interaction attribute features.

7. The method of claim 1, wherein the retrieving the resource according to the associated search information, obtaining the target resource comprises:

determining a target resource index according to the associated search information; and

the target resource associated with the target resource index is determined.

8. The method of claim 7, wherein the associated search information comprises a plurality of the associated search information associated with a same one of the target resource indices;

wherein said determining the target resource associated with the target resource index comprises:

a plurality of the target resources associated with the target resource index is determined.

9. The method of claim 8, wherein the target resource corresponds to the associated search information, the plurality of target resources being arranged based on recall order attributes corresponding to each of the plurality of associated search information;

the method further comprises the steps of:

and displaying the plurality of target resources based on the arrangement sequence attribute corresponding to each of the plurality of target resources.

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

obtaining a training sample, wherein the training sample comprises sample search information, sample preset search information and a sample label, the sample label comprises a semantic correlation label and an interaction attribute label, the semantic correlation label represents semantic correlation between the sample search information and the sample preset search information, and the interaction attribute label represents correlation between the sample interaction search information, the sample preset search information and historical retrieval interaction operation;

Inputting the sample search information and the sample preset search information into an initial deep learning model, and outputting sample semantic relevance characteristics and sample interaction attribute characteristics; and

training the initial deep learning model according to the sample semantic relevance features, the sample interaction attribute features, the semantic relevance labels and the interaction attribute labels to obtain a trained deep learning model.

11. The method of claim 10, wherein the initial deep learning model comprises a semantic feature extraction layer, a semantic relevance detection layer, and an interaction attribute detection layer;

the inputting the sample search information and the sample preset search information into an initial deep learning model, and outputting sample semantic relevance features and sample interaction attribute features comprises:

inputting the sample search information and the sample preset search information into the semantic feature extraction layer, and outputting sample search semantic features and sample preset search semantic features;

inputting the sample searching semantic features and the sample preset searching semantic features into the semantic relevance detection layer, and outputting the sample semantic relevance features; and

And inputting the sample searching semantic features and the sample preset searching semantic features into the interaction attribute detection layer, and outputting the sample interaction attribute features.

12. The method of claim 10, further comprising:

processing the sample search information and the sample preset search information based on a pre-trained semantic classification model to obtain semantic classification results between the sample search information and the sample preset search information; and

and determining the semantic relevance labels according to the semantic classification results.

13. The method of claim 10, wherein training the initial deep learning model based on the sample semantic relevance features, the sample interaction attribute features, the semantic relevance labels, and the interaction attribute labels comprises:

determining a first loss value according to the sample semantic relevance feature and the semantic relevance label;

determining a second loss value according to the sample interaction attribute characteristics and the interaction attribute labels; and

training the initial deep learning model according to the first loss value and the second loss value.

14. A resource retrieval device, comprising:

the detection module is used for responding to the received search information, carrying out semantic relevance detection and interaction attribute detection on the search information to obtain semantic relevance features and interaction attribute features, wherein the semantic relevance features represent semantic relevance between the search information and preset search information in a preset search information base, and the interaction attribute features represent relevance between the search information and search interaction operation;

The associated search information determining module is used for determining associated search information corresponding to the search information from the preset search information base according to the semantic relevance characteristics and the interaction attribute characteristics; and

and the resource retrieval module is used for retrieving the resources according to the associated search information to obtain target resources.

15. The apparatus of claim 14, wherein the association search information determination module comprises:

the search fusion feature obtaining sub-module is used for carrying out feature fusion on the semantic relevance features and the interaction attribute features to obtain search fusion features;

the searching information node determining submodule is used for determining searching information nodes according to the searching fusion characteristics;

the related search information node obtaining sub-module is used for carrying out neighbor search in a search information topological graph according to the search information node to obtain related search information nodes, wherein the search information topological graph comprises preset search nodes corresponding to preset search information and side relations among a plurality of preset search nodes, and the side relations represent semantic relativity among a plurality of preset search information; and

And the associated search information determining submodule is used for determining the associated search information according to the associated search information node.

16. The apparatus of claim 15, wherein the associated search information node obtaining submodule comprises:

the candidate association search information node obtaining unit is used for carrying out neighbor search in the search information topological graph according to the search information node to obtain a plurality of candidate association search information nodes;

the similarity detection result obtaining unit is used for carrying out semantic similarity detection on the search information and candidate associated search information associated with the candidate associated search information node to obtain a similarity detection result; and

and the association search information determining unit is used for determining the association search information from the candidate association search information according to the similarity detection result.

17. The apparatus of claim 15, wherein the search information topology map is constructed using:

18. The apparatus of claim 17, wherein the preset search information comprises a plurality of;

the method further comprises the following steps:

19. The apparatus of claim 14, wherein the detection module comprises:

the searching semantic feature obtaining sub-module is used for extracting semantic features of the searching information to obtain searching semantic features;

the semantic correlation feature obtaining sub-module is used for extracting semantic correlation features of the search semantic features to obtain the semantic correlation features; and

And the interaction attribute feature obtaining sub-module is used for extracting the interaction attribute features of the search semantic features to obtain the interaction attribute features.

20. The apparatus of claim 14, wherein the resource retrieval module comprises:

a target resource index determining sub-module for determining a target resource index according to the associated search information; and

a target resource determination sub-module for determining the target resource associated with the target resource index.

21. The apparatus of claim 20, wherein the associated search information comprises a plurality of the associated search information associated with a same one of the target resource indices;

wherein the target resource determination submodule includes:

and a target resource determining unit configured to determine a plurality of target resources associated with the target resource index.

22. The apparatus of claim 21, wherein the target resource corresponds to the associated search information, a plurality of the target resources being arranged based on recall order attributes corresponding to each of the plurality of associated search information;

the apparatus further comprises:

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

the training sample acquisition module is used for acquiring a training sample, wherein the training sample comprises sample search information, sample preset search information and a sample label, the sample label comprises a semantic correlation label and an interaction attribute label, the semantic correlation label represents semantic correlation between the sample search information and the sample preset search information, and the interaction attribute label represents correlation among the sample interaction search information, the sample preset search information and historical retrieval interaction operation;

the feature obtaining module is used for inputting the sample search information and the sample preset search information into an initial deep learning model and outputting sample semantic relevance features and sample interaction attribute features; and

the training module is used for training the initial deep learning model according to the sample semantic relevance characteristics, the sample interaction attribute characteristics, the semantic relevance labels and the interaction attribute labels to obtain a trained deep learning model.

24. The apparatus of claim 23, wherein the initial deep learning model comprises a semantic feature extraction layer, a semantic relevance detection layer, and an interaction attribute detection layer;

Wherein the feature obtaining module includes:

the first feature obtaining submodule is used for inputting the sample search information and the sample preset search information into the semantic feature extraction layer and outputting sample search semantic features and sample preset search semantic features;

the sample semantic relevance feature obtaining submodule is used for inputting the sample search semantic features and the sample preset search semantic features into the semantic relevance detection layer and outputting the sample semantic relevance features; and

and the sample interaction attribute feature obtaining submodule is used for inputting the sample search semantic features and the sample preset search semantic features into the interaction attribute detection layer and outputting the sample interaction attribute features.

25. The apparatus of claim 23, further comprising:

the semantic classification result obtaining module is used for processing the sample search information and the sample preset search information based on a pre-trained semantic classification model to obtain a semantic classification result between the sample search information and the sample preset search information; and

and the semantic relevance label determining module is used for determining the semantic relevance label according to the semantic classification result.

26. The apparatus of claim 23, wherein the training module comprises:

a second loss value determining submodule, configured to determine a second loss value according to the sample interaction attribute feature and the interaction attribute tag; and

27. An electronic device, comprising:

at least one processor; and

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

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 13.

28. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 13.

29. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 13.