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

Abstract

The application relates to the technical field of information retrieval, and in particular to a retrieval result determination method and device, electronic equipment and a storage medium. The method comprises the following steps: determining first association feature information of a target retrieval word and a candidate object based on semantic information of the target retrieval word and semantic information of the candidate object; inputting the first association feature information into a matching data prediction network to obtain candidate matching data of the target retrieval word and the candidate object; the matching data prediction network is obtained by training based on data matching samples containing sample label data; the sample label data is obtained by performing information matching on sample retrieval words and sample objects; and a plurality of target retrieval objects corresponding to the target retrieval word are determined from the candidate object based on the candidate matching data. The application can improve the accuracy of sample construction and the accuracy of retrieval results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information retrieval, and in particular to a retrieval result determination method and device, electronic equipment and a storage medium. BACKGROUND

[0002] Information retrieval refers to a process of retrieving results matched with user intentions according to keywords. Retrieval is applied in different scenarios, such as Internet retrieval, paper retrieval, e-commerce retrieval, etc.

[0003] For example, in the early stage of the e-commerce industry, due to less user operation data and low quality of the user operation data, constructing a sample based on the user operation data will lead to inaccurate sample construction, and thus constructing a model based on the sample will lead to inaccurate retrieval results. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a retrieval result determination method and device, electronic equipment and a storage medium, which can improve the accuracy of sample construction and thus improve the accuracy of retrieval results.

[0005] To solve the above technical problem, on the one hand, a retrieval result determination method is provided, comprising:

[0006] determining first association feature information of a target keyword and a candidate object based on semantic information of the target keyword and semantic information of the candidate object;

[0007] inputting the first association feature information into a matching data prediction network to obtain candidate matching data of the target keyword and the candidate object; the matching data prediction network is trained based on data matching samples containing sample label data; the sample label data is obtained by information matching of sample keywords and sample objects;

[0008] determining a plurality of target retrieval objects corresponding to the target keyword from the candidate objects based on the candidate matching data.

[0009] On the other hand, a retrieval result determination device is provided, comprising:

[0010] a first association feature determination module configured to determine first association feature information of a target keyword and a candidate object based on semantic information of the target keyword and semantic information of the candidate object;

[0011] The matching data prediction module is configured to input the first associated feature information into a matching data prediction network to obtain candidate matching data of the target search term and the candidate object, wherein the matching data prediction network is trained based on data matching samples containing sample label data, and the sample label data is obtained by performing information matching on sample search terms and sample objects.

[0012] The search result determination module is configured to determine a plurality of target search objects corresponding to the target search term from the candidate objects based on the candidate matching data.

[0013] In another aspect, the present application provides an electronic device, which comprises a processor and a memory, and the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the search result determination method as described above.

[0014] In another aspect, the present application provides a computer storage medium, which stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the search result determination method as described above.

[0015] The implementation of the embodiments of the present application has the following beneficial effects:

[0016] The present application determines the first associated feature information of the target search term and the candidate object based on the semantic information of the target search term and the semantic information of the candidate object, inputs the first associated feature information into a matching data prediction network to obtain candidate matching data of the target search term and the candidate object, wherein the matching data prediction network is trained based on data matching samples containing sample label data, and the sample label data is obtained by performing information matching on sample search terms and sample objects, and then determines a plurality of target search objects corresponding to the target search term from the candidate objects based on the candidate matching data. That is, the sample label is obtained by performing information matching on sample search terms and sample objects, which is different from the method of determining the sample label based on user operation data, thereby avoiding the problem of inaccurate sample caused by low quality of user operation data and improving the accuracy of sample construction. The matching data prediction network is trained based on the constructed sample, which can improve the prediction accuracy of the matching data prediction model, and then in the information search process, the candidate matching data is predicted based on the matching data prediction network, and the target search object is determined from the candidate objects based on the candidate matching data, which can improve the accuracy of the search result. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, below will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 is an implementation environment schematic diagram provided by the embodiments of the present application;

[0019] Figure 2 is a retrieval result determination method flowchart provided by the embodiments of the present application;

[0020] Figure 3 is a method flowchart for generating sample label data based on information matching provided by the embodiments of the present application;

[0021] Figure 4 is a method flowchart for determining dimension matching information of each matching dimension provided by the embodiments of the present application;

[0022] Figure 5 is a method flowchart for determining sub-matching information of multiple sub-matching dimensions provided by the embodiments of the present application;

[0023] Figure 6 is a method flowchart for obtaining sample label data based on weighted calculation provided by the embodiments of the present application;

[0024] Figure 7 is a matching data prediction network training method flowchart provided by the embodiments of the present application;

[0025] Figure 8 is a method flowchart for constructing associated feature information provided by the embodiments of the present application;

[0026] Figure 9 is an implementation schematic diagram of a matching data prediction network provided by the embodiments of the present application;

[0027] Figure 10 is a retrieval result determination device schematic diagram provided by the embodiments of the present application;

[0028] Figure 11 is an electronic device structure schematic diagram provided by the embodiments of the present application. DETAILED DESCRIPTION

[0029] In order to make the objects, technical solutions and advantages of the present application clearer, the following further describes the present application with reference to the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0030] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in other than the order illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units need not be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0031] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0032] Please refer to Figure 1 which shows an implementation environment provided by the embodiments of the present application, which can include at least one first terminal 110 and a second terminal 120, and the first terminal 110 and the second terminal 120 can communicate data through a network.

[0033] Specifically, the first terminal 110 sends an information retrieval request to the second terminal 120, and the information retrieval request includes a target search term; the second terminal 120 receives the information retrieval request, performs information matching with the candidate object based on the target search term, determines the target search object from the candidate object based on the information matching result, generates a search result based on the target search object and returns it to the first terminal 110.

[0034] The first terminal 110 can communicate with the second terminal 120 based on a browser / server mode (B / S) or a client / server mode (C / S). The first terminal 110 can include an entity device such as a smartphone, a tablet computer, a notebook computer, a digital assistant, a smart wearable device, a vehicle-mounted terminal, a server, and the like, or a software such as an application program running in the entity device. The operating system running on the first terminal 110 in the embodiments of the present application can include, but is not limited to, an Android system, an IOS system, Linux, Windows, and the like.

[0035] The second terminal 120 can establish a communication connection with the first terminal 110 through a wire or wireless. The second terminal 120 can include a server running independently, or a distributed server, or a server cluster composed of multiple servers, wherein the server can be a cloud server.

[0036] In order to solve the problem that the sample constructed based on the user operation data in the prior art can lead to inaccurate sample construction, and then the model constructed based on the sample can lead to inaccurate retrieval results, the embodiments of the present application provide a retrieval result determination method. The execution subject of the method can be the second terminal described above. Specifically, please refer to Figure 2 The method can include:

[0037] S210. Based on the semantic information of the target search term and the semantic information of the candidate object, determine the first association feature information of the target search term and the candidate object.

[0038] In the embodiments, the target search term can include one or more keywords. When multiple keywords are included, the multiple keywords can be connected by a preset search symbol to obtain a corresponding target search term. The preset search symbol can include, but is not limited to, and, or, not, and the like. The search symbols in the prior art can be applied to the embodiments.

[0039] Specifically, the target search term can include keywords in multiple languages. When multiple keywords are included in the target keyword, each keyword in the target search term can be in the same language or in different languages. For example, the target search term can be brand A, milk, or the target search term can be brand A, milk.

[0040] The semantic information of the target search term can be obtained by performing semantic analysis on the target keyword. The semantic analysis can specifically include steps such as word segmentation processing and semantic recognition. The semantic analysis process can be implemented by using a preset semantic analysis model. The semantic analysis model can perform word segmentation on the input target search term. During the word segmentation process, a corresponding word segmentation method can be used according to the grammar of different languages. Then, semantic recognition can be performed on the word segmentation. For sample search terms containing different languages, the semantic recognition can specifically further include word segmentation translation, so as to translate the word segmentation of multiple languages into the word segmentation of the target language. This facilitates subsequent information matching and improves the information matching efficiency.

[0041] The semantic information of the candidate object can be obtained by using a semantic analysis model in advance and stored. Therefore, the semantic information of the candidate object can be analyzed only once, and repeated semantic analysis is avoided. This can improve the analysis efficiency of the semantic information of the candidate object and save system computing resources.

[0042] The first associated feature information can represent the matching feature information of the target search term and the candidate object. The first associated feature information can be determined by the matching result of the semantic information of the target search term and the semantic information of the candidate object. The first associated feature information can be a feature sequence or a feature vector, and each dimension of the feature sequence or the feature vector can be used to represent the matching feature information of the target search term and the candidate object.

[0043] In S220, the first associated feature information is input into a matching data prediction network to obtain candidate matching data of the target search term and the candidate object. The matching data prediction network is trained based on data matching samples containing sample label data. The sample label data is obtained by performing information matching on sample search terms and sample objects.

[0044] The matching data prediction network can predict the matching degree of the target keyword and the candidate object based on the associated feature information of the input target search term and the candidate object, and then obtain the candidate matching data of the target search term and the candidate object. The candidate matching data can be a numerical value or a degree identifier. The matching degree of the target keyword and the candidate object can be determined by the candidate matching data.

[0045] The matching data prediction network can be trained based on data matching samples containing sample label data. The sample label data in the embodiments of the present application is obtained by performing information matching on sample search terms and sample objects. For different sample search term-sample object matching pairs, matching can be performed respectively to obtain corresponding sample label data. Therefore, the obtained sample label data can represent the information of the corresponding sample search term-sample object matching pair, thereby avoiding the problem of inaccurate sample label data caused by directly using user operation data to construct samples.

[0046] In addition, the sample search terms can be selected from the user operation data or randomly selected, and the sample objects can be selected from the recall results or randomly selected, and the embodiment is not limited specifically.

[0047] S230. Based on the candidate matching data, a plurality of target search objects corresponding to the target search terms are determined from the candidate objects.

[0048] According to the above content, the candidate matching data can represent the matching degree of the target keyword and the candidate object, so that when the search result is determined, the candidate object with a higher matching degree with the target search term can be selected from the candidate objects as the target search object.

[0049] Specifically, when the target search object is determined, the candidate object with the candidate matching data greater than or equal to the preset matching data can be determined as the target search object, or the N candidate objects ranked in the top N based on the candidate matching data from large to small are determined as the target search object.

[0050] The embodiment of the application obtains the sample label by information matching of the sample search term and the sample object, which is different from the method of determining the sample label based on the user operation data, avoids the problem of inaccurate sample caused by low quality of the user operation data, and improves the accuracy of sample construction. The matching data prediction network is trained based on the constructed sample, which can improve the prediction accuracy of the matching data prediction model, and then in the information retrieval process, the candidate matching data is predicted based on the matching data prediction network, and the target search object is determined from the candidate object based on the candidate matching data, which can improve the accuracy of the search result.

[0051] Please refer to Figure 3 which shows a method for generating sample label data based on information matching, which can include:

[0052] S310. Determine the semantic information of a plurality of matching dimensions of the sample search term, and the semantic information of a plurality of matching dimensions of the sample object.

[0053] S320. Match the semantic information of the sample search term and the sample object in the plurality of matching dimensions respectively, to obtain dimension matching information corresponding to the plurality of matching dimensions respectively.

[0054] S330. Based on the dimension matching information corresponding to the plurality of matching dimensions respectively, the sample label data is obtained.

[0055] The semantic information of the sample search term in multiple matching dimensions can be extracted from the semantic information of the sample search term, and the semantic information of the sample object in multiple matching dimensions can be extracted from the semantic information of the sample object in advance and stored.

[0056] In the process of generating the sample label data, the semantic information of the sample search term and the sample object in multiple matching dimensions can be matched respectively; the semantic information under each matching dimension can represent the characteristic information of the matching dimension, and the matching of the sample search term and the semantic information of the sample object in the same matching dimension can obtain the matching information of the sample search term and the sample object in the matching dimension.

[0057] Taking e-commerce search as an example, the object can be specifically a commodity, so the multiple matching dimensions can specifically include a category matching dimension, a brand matching dimension, a product word matching dimension, and a commodity description matching dimension; by matching the semantic information in the category matching dimension, the category matching information of the sample search term and the sample commodity can be determined; by matching the semantic information in the brand matching dimension, the brand matching information of the sample search term and the sample commodity can be determined; by matching the semantic information in the product word matching dimension, the product word matching information of the sample search term and the sample commodity can be determined; and by matching the semantic information in the commodity description matching dimension, the commodity description matching information of the sample search term and the sample commodity can be determined.

[0058] Therefore, when the sample search term and the sample object are matched, the semantic information in multiple matching dimensions can be matched respectively, and the sample label data can be determined based on the matching results of the multiple matching dimensions, so that comprehensive matching of the sample search term and the sample object in multiple matching dimensions can be realized, and the comprehensiveness of the matching of the sample search term and the sample object and the accuracy of the determination of the sample label data can be improved.

[0059] Further, each matching dimension includes multiple sub-matching dimensions; correspondingly, please refer to Figure 4 which shows a method for determining the dimension matching information of each matching dimension, which can include:

[0060] S410. In each matching dimension, the semantic information of the sample search term and the sample object in the multiple sub-matching dimensions is matched respectively to obtain sub-matching information corresponding to the multiple sub-matching dimensions respectively.

[0061] S420. The sub-matching information corresponding to the multiple sub-matching dimensions respectively is summed to obtain the dimension matching information of each matching dimension.

[0062] Specifically, each matching dimension can further include a plurality of sub-matching dimensions, so that when the semantic information under each matching dimension is matched, the semantic information of the plurality of sub-matching dimensions under the matching dimension can be matched to obtain sub-matching information corresponding to the plurality of sub-matching dimensions respectively.

[0063] Taking the above category matching dimension as an example, the category matching dimension can include a plurality of levels of category matching dimensions, i.e., each level of category corresponds to a sub-matching dimension. The category can include a first-level category, a second-level category, a third-level category, a fourth-level category, and so on, and each level of category corresponds to a sub-matching dimension. Specifically, if the sample search term is “cola”, it can be determined that the first-level category of the sample search term is beverage, the second-level category is carbonated beverage, and the third-level category is cola. Correspondingly, the multi-level category information corresponding to the sample search term and the multi-level category information of the sample object can be matched respectively to obtain matching information in each sub-matching dimension. Then, the matching information under each sub-matching dimension is summed to obtain the dimension matching information under the matching dimension.

[0064] Taking the above brand matching dimension as an example, the brand matching dimension can include a plurality of language brand matching dimensions, and each language brand matching dimension corresponds to a sub-matching dimension. For example, for brand A, the Chinese expression is “brand A”, and the English expression is “A”. Correspondingly, the multi-language brand matching dimension corresponding to the sample search term and the semantic information under the multi-language brand matching dimension of the sample object can be matched respectively to obtain matching information in each sub-matching dimension. Then, the matching information under each sub-matching dimension is summed to obtain the dimension matching information under the matching dimension.

[0065] Taking the above product word matching dimension as an example, the product word matching dimension can include a plurality of language product word matching dimensions, and each language product word matching dimension corresponds to a sub-matching dimension. For example, for the product word “cola”, the Chinese expression is “cola”, and the English expression is “coke”. Correspondingly, the multi-language product word matching dimension corresponding to the sample search term and the semantic information under the multi-language product word matching dimension of the sample object can be matched respectively to obtain matching information in each sub-matching dimension. Then, the matching information under each sub-matching dimension is summed to obtain the dimension matching information under the matching dimension.

[0066] Taking the above product description matching dimension as an example, the product description generally can include attribute information of the product such as efficacy, characteristics, specifications, and so on, so that each attribute information dimension can correspond to a sub-matching dimension. Correspondingly, the plurality of attribute information dimensions corresponding to the sample search term and the semantic information under the plurality of attribute information dimensions of the sample object can be matched respectively to obtain matching information in each sub-matching dimension. Then, the matching information under each sub-matching dimension is summed to obtain the dimension matching information under the matching dimension.

[0067] Specifically, the calculation method of the matching information corresponding to one matching dimension is shown as formula (1):

[0068]

[0069] wherein, I(x) is an indicator function:

[0070]

[0071] Q i for each sub-matching dimension corresponding to the sample search term, G j for each sub-matching dimension corresponding to the sample object.

[0072] Therefore, a plurality of sub-matching dimensions are further included under each matching dimension, by respectively matching the semantic information under the sub-matching dimension under each matching dimension, a plurality of sub-matching information respectively corresponding to the sub-matching dimensions are obtained, and the plurality of sub-matching information respectively corresponding to the sub-matching dimensions are summed to obtain the dimension matching information of each matching dimension, thereby on the one hand, comprehensive matching of the sample search term and the sample object under each matching dimension can be realized; on the other hand, the problem of matching failure caused by not setting the multi-lingual sub-matching dimension can be avoided, thereby the problem of inaccurate matching information is further avoided, and the accuracy of the sample label data determination is further improved.

[0073] Please refer to Figure 5 which shows a determination method of the sub-matching information of the plurality of sub-matching dimensions, which can include:

[0074] S510. Under each sub-matching dimension, it is judged whether the semantic information of the sample search term exists corresponding semantic information of the sub-matching dimension; if yes, step S520 is executed; if no, step S530 is executed.

[0075] S520. The semantic information of the sample search term in the sub-matching dimension is matched with the semantic information of the sample object in the sub-matching dimension to obtain the sub-matching information of the sample search term and the sample object in the sub-matching dimension.

[0076] S530. It is determined that the preset matching information is the sub-matching information.

[0077] Taking the brand matching dimension as an example, the brand matching dimension can include a multi-lingual brand matching dimension, and each language brand matching dimension corresponds to a sub-matching dimension. For example, for brand A, the Chinese expression is "brand A", and the English expression is "A". The sample search term contains "brand A", that is, the semantic information of the sample search term exists in the semantic information of the Chinese brand matching dimension, and then the "brand A" in the sample search term can be matched with the semantic information of the sample object in the Chinese brand matching dimension to obtain the sub-matching information of "brand A" in the Chinese brand matching dimension. The sample search term does not contain "A", that is, the semantic information corresponding to the English brand matching dimension does not exist in the semantic information of the sample search term, and the preset matching information can be determined as the corresponding sub-matching information. The preset matching information can be "NULL" or "0".

[0078] Therefore, when performing semantic information matching of the sub-matching dimension, it can be first judged whether the sample search term exists in each sub-matching dimension. If it exists, the corresponding matching is performed to obtain the sub-matching information. If it does not exist, the preset matching information is determined as the sub-matching information, thereby improving the efficiency of information matching.

[0079] In addition, please refer to Figure 6 which shows a method for obtaining sample label data based on weighted calculation, which can include:

[0080] S610. Determine the matching weight corresponding to the dimension matching information of each matching dimension.

[0081] S620. Based on the matching weight, the dimension matching information of the plurality of matching dimensions is weighted and summed to obtain the sample label data.

[0082] The matching weight corresponding to each matching dimension can be set based on different scene needs, so that on the basis of calculating the dimension matching information corresponding to each matching dimension, the weighted sum of each dimension matching information is obtained to obtain the corresponding sample label data. Since the matching weight is set based on different scene needs, the calculated sample label data is adapted to the application scene, and the matching weight can be flexibly set, thereby improving the flexibility of sample label data determination.

[0083] The specific calculation method of the sample label data is shown in formula (3):

[0084]

[0085] Wherein, a, b, g, d are matching weights. Specifically, the matching weight values can be: a = 0.4, b = 0.25, g = 0.2, d = 0.15.

[0086] Please refer to Figure 7 which shows a matching data prediction network training method, which can include:

[0087] S710. Based on the semantic information of the sample search term and the semantic information of the sample object, determine the second association feature information of the sample search term and the sample object.

[0088] S720. Input the second association feature information into the preset data prediction network to obtain the predicted matching data of the sample search term and the sample object.

[0089] S730. Based on the sample label data and the predicted matching data, train the preset data prediction network to obtain the matching data prediction network.

[0090] The determination method of the second association feature information is similar to that of the first association feature information, which will not be repeated here.

[0091] By inputting the second association feature information into the preset data prediction network for prediction, the predicted matching data is obtained; based on the sample label data and the predicted matching data, the parameters of the preset data prediction network are adjusted to obtain the matching data prediction network. The matching data prediction network in this embodiment can be a regression model, which can include traditional machine learning models such as lr (Linear Regression, linear regression), xgboost Regression, and deep learning regression models; the prediction value of the matching data prediction network is a continuous value.

[0092] Since the sample label data is obtained by matching the information of the sample search term and the sample object, it can be obtained by matching the semantic information of multiple matching dimensions of the sample search term and the semantic information of multiple matching dimensions of the sample object, thereby improving the accuracy of the sample label data; further based on the sample label data for model training, the accuracy of the matching data prediction network prediction can be improved.

[0093] For the specific construction method of the association feature information, please refer to Figure 8 which can include:

[0094] S810. Based on the semantic information of the target search term and the semantic information of the candidate object, determine the matching features of the target search term and the candidate object in multiple feature dimensions.

[0095] S820. Based on the matching features in the multiple feature dimensions, construct the features to obtain the first association feature information.

[0096] Specifically, a plurality of matching feature information items can be pre-set; the matching feature information of the target search word and the candidate object in each feature information item is obtained by matching the semantic information of the target search word and the semantic information of the candidate object. For example, the category matching feature information item, the brand word matching feature information item, the product word matching feature information item, the search word and the product title matching feature item, etc. can be pre-set, each matching feature information item can be represented by one feature dimension or a plurality of feature dimensions, and the specific feature value of each feature dimension can be a decimal between 0 and 1, or 0 or 1. The first associated feature information is obtained by sequentially splicing the feature values of the plurality of feature dimensions, and is specifically applied in the form of a feature vector or a feature sequence.

[0097] For the category matching feature information item, it can be judged whether the multi-level category of the target search word and the candidate object matches, and if it matches, it is 1, and if it does not match, it is 0; for the brand word matching feature information item, it can be judged whether the brand of the target search word and the candidate object matches, and if it matches, it is 1, and here the matching includes matching consistency and alias matching, and if it does not match, it is 0; for the product word matching feature information item, the number of product words in the target search word that match or are the same as the candidate object is calculated, and the proportion of the target search word segmentation is calculated.

[0098] By matching the semantic information of the target search word and the semantic information of the candidate object, the matching feature in the plurality of feature dimensions can be obtained, which can improve the convenience and efficiency of matching feature determination; further, the feature construction is performed based on the matching features in the plurality of feature dimensions to obtain the associated feature information, and each feature value of each feature dimension has corresponding feature information, thereby improving the explainability of the associated feature information.

[0099] The above-mentioned associated feature information can also be determined by a deep network model, and the specific steps can include: inputting the semantic information of the target search word and the semantic information of the candidate object into the deep network model to obtain a feature word vector of the target search word and the candidate object, and the feature word vector can be used as an associated feature vector.

[0100] Specifically, please refer to Figure 9 When training the matching data prediction network, first, the matching features of the <query, product> sample are extracted, and then the extracted matching features are input into the matching data prediction network to obtain the corresponding predicted matching data; when predicting the matching data based on the matching data prediction network, first, the matching features of the <target search word, product> are extracted, and then the extracted matching features are input into the matching data prediction network to obtain the corresponding predicted matching data.

[0101] The retrieval result determination method provided by the embodiment can be applied to the field of e-commerce retrieval or the field of paper retrieval, and can be applied to the early stage of a project, a scenario in which user operation data is less and data quality is not high, and the like. The method can improve retrieval effect and user experience.

[0102] The embodiment further provides a retrieval result determination device, which can be used for determining a retrieval result. Figure 10 The device can include:

[0103] The first association feature determination module 1010 is configured to determine first association feature information of the target retrieval word and the candidate object based on semantic information of the target retrieval word and semantic information of the candidate object.

[0104] The matching data prediction module 1020 is configured to input the first association feature information into a matching data prediction network to obtain candidate matching data of the target retrieval word and the candidate object. The matching data prediction network is trained based on data matching samples containing sample label data.

[0105] The retrieval result determination module 1030 is configured to determine a plurality of target retrieval objects corresponding to the target retrieval word from the candidate object based on the candidate matching data.

[0106] Further, the device further includes:

[0107] The first determination module is configured to determine semantic information of a plurality of matching dimensions of the sample retrieval word and semantic information of a plurality of matching dimensions of the sample object.

[0108] The first matching module is configured to match the sample retrieval word and the sample object in the semantic information of the plurality of matching dimensions respectively to obtain dimension matching information corresponding to the plurality of matching dimensions respectively.

[0109] The sample label data generation module is configured to obtain the sample label data based on the dimension matching information corresponding to the plurality of matching dimensions respectively.

[0110] Further, the device further includes:

[0111] The second association feature information determination module is configured to determine second association feature information of the sample retrieval word and the sample object based on semantic information of the sample retrieval word and semantic information of the sample object.

[0112] The first prediction module is configured to input the second association feature information into a preset data prediction network to obtain predicted matching data of the sample retrieval word and the sample object.

[0113] a training module configured to train the preset data prediction network based on the sample label data and the predicted matching data, to obtain the matching data prediction network.

[0114] Further, each matching dimension includes a plurality of sub-matching dimensions; the first matching module includes:

[0115] a second matching module configured to match the semantic information of the sample search term and the sample object in the plurality of sub-matching dimensions under each matching dimension, to obtain sub-matching information corresponding to the plurality of sub-matching dimensions respectively;

[0116] a summation module configured to sum the sub-matching information corresponding to the plurality of sub-matching dimensions respectively, to obtain dimension matching information of each matching dimension.

[0117] Further, the second matching module includes:

[0118] a second determination module configured to, under each sub-matching dimension, if the semantic information of the sample search term exists corresponding semantic information of the sub-matching dimension, match the semantic information of the sample search term in the sub-matching dimension with the semantic information of the sample object in the sub-matching dimension, to obtain sub-matching information of the sample search term and the sample object in the sub-matching dimension;

[0119] a third determination module configured to, if the semantic information of the sample search term does not exist corresponding semantic information of the sub-matching dimension, determine preset matching information as the sub-matching information

[0120] Further, the sample label data generation module includes:

[0121] a matching weight determination module configured to determine matching weights corresponding to the dimension matching information of the plurality of matching dimensions respectively;

[0122] a weighted summation module configured to perform weighted summation on the dimension matching information of the plurality of matching dimensions based on the matching weights, to obtain the sample label data.

[0123] Further, the first association feature determination module includes:

[0124] a fourth determination module configured to determine matching features of the target search term and the candidate object in a plurality of feature dimensions based on the semantic information of the target search term and the semantic information of the candidate object;

[0125] a feature construction module configured to perform feature construction based on the matching features in the plurality of feature dimensions, to obtain the first association feature information.

[0126] The apparatus provided in the above embodiments can execute the method provided in any of the embodiments of the present application, and has the corresponding function modules and advantages of executing the method. Technical details not described in the above embodiments can be referred to the method provided in any of the embodiments of the present application.

[0127] The embodiment further provides a computer readable storage medium, wherein the storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by a processor to perform any of the above methods.

[0128] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device perform any of the above methods.

[0129] Figure 11 is a block diagram of an electronic device for a search result determination method according to an exemplary embodiment. The electronic device can be a server, and its internal structure diagram can be as shown in Figure 11 The electronic device includes a processor, a memory and a network interface connected through a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the electronic device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a search result determination method.

[0130] The present specification provides method operation steps as described in the embodiments or flowcharts, but more or fewer operation steps can be included based on conventional or non-inventive labor. The steps and order listed in the embodiments are only one of the many execution orders, and do not represent the only execution order. When the system or interrupt product is executed in practice, it can be executed in sequence or in parallel (such as parallel processor or multi-threaded environment) according to the method order shown in the embodiments or drawings.

[0131] The structure shown in the embodiments only relates to part of the structure of the application, and does not constitute a limitation on the device to which the application is applied. The specific device can include more or fewer components than shown, or combine certain components, or have a different arrangement of components. It should be understood that the methods, devices, etc. disclosed in the embodiments can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. In actual implementation, another division mode can be used, for example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, devices or unit modules.

[0132] Based on such understanding, the technical solutions of the application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0133] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in the specification can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solutions. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0134] The above embodiments are only used to illustrate the technical solutions of the application, but not to limit it. Although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features. Such modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A search result determination method characterized by comprising: The method comprises the following steps: determining first association feature information of the target search term and the candidate object based on semantic information of the target search term and semantic information of the candidate object; inputting the first association feature information into a matching data prediction network to obtain candidate matching data of the target search term and the candidate object; the matching data prediction network is trained based on data matching samples containing sample label data; the sample label data is obtained by information matching of sample search terms and sample objects; the determination method of the sample label data comprises: determining semantic information of multiple matching dimensions of the sample search term and semantic information of multiple matching dimensions of the sample object; at least one matching dimension comprises multiple language sub-matching dimensions; each matching dimension comprises multiple sub-matching dimensions; under each sub-matching dimension, if the semantic information of the sample search term has corresponding semantic information of the sub-matching dimension, the semantic information of the sample search term in the sub-matching dimension is matched with the semantic information of the sample object in the sub-matching dimension to obtain sub-matching information of the sample search term and the sample object in the sub-matching dimension; the sample search term in the multiple language sub-matching dimensions corresponding to the at least one matching dimension is matched with the sample object in the multiple language sub-matching dimensions corresponding to the at least one matching dimension to obtain dimension matching information under each language sub-matching dimension; the dimension matching information under each language sub-matching dimension is summed to obtain dimension matching information under the at least one matching dimension; if the semantic information of the sample search term does not have corresponding semantic information of the sub-matching dimension, a preset matching information is determined as the sub-matching information; the sub-matching information corresponding to the multiple sub-matching dimensions is summed to obtain dimension matching information of each matching dimension; based on the dimension matching information corresponding to the multiple matching dimensions, the sample label data is obtained; based on the candidate matching data, multiple target search objects corresponding to the target search term are determined from the candidate objects.

2. The method of claim 1, wherein, The method further comprises: determining second association feature information of the sample search term and the sample object based on semantic information of the sample search term and semantic information of the sample object; inputting the second association feature information into a preset data prediction network to obtain predicted matching data of the sample search term and the sample object; training the preset data prediction network based on the sample label data and the predicted matching data to obtain the matching data prediction network.

3. The method of claim 1, wherein, The method of obtaining the sample label data based on the dimension matching information corresponding to the multiple matching dimensions comprises: determining matching weights corresponding to the dimension matching information of the multiple matching dimensions; based on the matching weights, the dimension matching information of the multiple matching dimensions is weighted and summed to obtain the sample label data.

4. The method of claim 1, wherein, The semantic information of the target search word and the semantic information of the candidate object are used to determine first association feature information of the target search word and the candidate object, which includes: Based on the semantic information of the target search word and the semantic information of the candidate object, the matching features of the target search word and the candidate object in multiple feature dimensions are determined. Based on the matching features in multiple feature dimensions, feature construction is performed to obtain the first association feature information.

5. A search result determination apparatus characterized by comprising: It includes: A first association feature determination module is configured to determine first association feature information of a target search word and a candidate object based on semantic information of the target search word and semantic information of the candidate object. A matching data prediction module is configured to input the first association feature information into a matching data prediction network to obtain candidate matching data of the target search word and the candidate object. The sample label data is obtained by matching information of a sample search word and a sample object. The determination method of the sample label data includes determining semantic information of multiple matching dimensions of the sample search word and semantic information of multiple matching dimensions of the sample object. At least one matching dimension includes multiple language sub-matching dimensions, and each matching dimension includes multiple sub-matching dimensions. If the semantic information of the sample search word exists corresponding semantic information of the sub-matching dimension, the semantic information of the sample search word in the sub-matching dimension is matched with the semantic information of the sample object in the sub-matching dimension to obtain sub-matching information of the sample search word and the sample object in the sub-matching dimension.

6. The apparatus of claim 5, wherein, The semantic information of the sample search word in the at least one matching dimension corresponding to the multiple language sub-matching dimensions is matched with the semantic information of the sample object in the at least one matching dimension corresponding to the multiple language sub-matching dimensions to obtain dimension matching information in each language sub-matching dimension. The dimension matching information in each language sub-matching dimension is summed to obtain dimension matching information in the at least one matching dimension. If the semantic information of the sample search word does not exist corresponding semantic information of the sub-matching dimension, preset matching information is determined as the sub-matching information. The sub-matching information corresponding to the multiple sub-matching dimensions is summed to obtain dimension matching information of each matching dimension. Based on the dimension matching information corresponding to the multiple matching dimensions, the sample label data is obtained. A search result determination module is configured to determine multiple target search objects corresponding to the target search word from the candidate object based on the candidate matching data. The device further includes: A second association feature information determination module is configured to determine second association feature information of a sample search word and a sample object based on semantic information of the sample search word and semantic information of the sample object. A first prediction module is configured to input the second association feature information into a preset data prediction network to obtain predicted matching data of the sample search word and the sample object. The training module is configured to train the preset data prediction network based on the sample label data and the predicted matching data, and obtain the matching data prediction network.

7. The apparatus of claim 5, wherein, The sample label data generation module comprises: The matching weight determination module is configured to determine matching weights corresponding to the dimension matching information of the plurality of matching dimensions respectively. The weighted summation module is configured to perform weighted summation on the dimension matching information of the plurality of matching dimensions based on the matching weights, and obtain the sample label data.

8. The apparatus of claim 5, wherein, The first association feature determination module comprises: The fourth determination module is configured to determine matching features of the target search term and the candidate object in a plurality of feature dimensions based on the semantic information of the target search term and the semantic information of the candidate object. The feature construction module is configured to perform feature construction based on the matching features in the plurality of feature dimensions, and obtain the first association feature information.

9. An electronic device, comprising: The device comprises a processor and a memory, and the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the search result determination method according to any one of claims 1 to 4.

10. A computer storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the search result determination method according to any one of claims 1 to 4.

11. A computer program product, characterised in that, The computer program product comprises computer instructions stored in a computer readable storage medium; the processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the search result determination method according to any one of claims 1 to 4.

Images