Multi-recall sorting analysis method and system based on semantic retrieval

Abstract

The embodiment of the invention provides a multi-recall sorting analysis method and system based on semantic retrieval, and belongs to the technical field of big data services. The method comprises the following steps: acquiring a keyword input by a user; obtaining a recall result of each path according to the keyword; obtaining interest characteristics of the user; constructing a first sensitive tree of the user according to the interest characteristics; acquiring identity characteristics of the user; acquiring a second sensitive tree corresponding to the identity feature; fusing the first sensitive tree and the second sensitive tree to obtain a comprehensive sensitive tree of the user; and sorting the recall results according to the comprehensive sensitive tree. According to the method and the system, the multi-path recall results are sorted by combining the interest characteristics of the user and the identity characteristics in the system, so that the improvement of the retrieval method based on the multi-path recall is realized, and the retrieval accuracy and efficiency in the power system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data services, and in particular to a sorting and analysis method and system for multi-way recall based on semantic retrieval. Background Art

[0002] Multi-way recall is one of the common methods of search engines. It mainly performs semantic analysis based on keywords or sentences input by users and recalls documents through data in multiple dimensions to obtain accurate search results.

[0003] Existing multi-channel recall methods mainly rely on big data analysis, which can achieve accurate feedback when the database system is large and the number of samples is sufficient.

[0004] The power information communication management system is a closed system used to manage power grid data within the power grid. When facing user searches, the commonly used multi-channel recall method is difficult to adapt due to the lack of the necessary conditions (huge data volume and sufficient sample number) that the commonly used multi-channel recall method relies on. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a multi-way recall sorting analysis method and system based on semantic retrieval, which can meet the retrieval accuracy requirements within the power system.

[0006] In order to achieve the above object, an embodiment of the present invention provides a sorting analysis method based on semantic retrieval and multi-way recall, comprising:

[0007] Get the keywords entered by the user;

[0008] Obtaining the recall result of each route according to the keyword;

[0009] Acquire interest characteristics of the user;

[0010] Constructing a first sensitivity tree of the user according to the interest feature;

[0011] Obtaining identity characteristics of the user;

[0012] Obtaining a second sensitive tree corresponding to the identity feature;

[0013] Merging the first sensitivity tree and the second sensitivity tree to obtain a comprehensive sensitivity tree for the user;

[0014] The recall results are sorted according to the comprehensive sensitivity tree.

[0015] Optionally, obtaining the recall result of each path according to the keyword includes:

[0016] Performing an expansion operation on the keyword to obtain multiple synonyms;

[0017] Perform matching search in a database according to the keywords and synonyms to obtain the recall result;

[0018] Encoding the keywords and synonyms according to a preset word library to obtain corresponding keyword codes;

[0019] Inputting the keyword encoding into a preset language model to obtain a corresponding vector representation;

[0020] A similarity query is performed in the database according to the vector representation to obtain the recall result.

[0021] Optionally, obtaining the interest characteristics of the user includes:

[0022] Get the user's historical query records;

[0023] Classify and count the historical query records to determine the count value of each underlying index classification corresponding to the historical query records;

[0024] Performing statistics on the underlying index classification to determine the concentration of the underlying index classification;

[0025] Determining whether the concentration is greater than or equal to a preset concentration threshold;

[0026] In the case where it is determined that the concentration is greater than or equal to the concentration threshold, classifying the underlying index having a count value greater than a preset count value threshold as the interest feature;

[0027] In the case where it is determined that the concentration is less than the concentration threshold, dividing the historical query record into a plurality of sub-records according to a preset time interval;

[0028] Traversing each of the sub-records to determine the count value of the underlying index classification included in each of the sub-records;

[0029] updating each of the count values ​​according to the distance from the current time to obtain a comprehensive count value of the underlying index classification;

[0030] The comprehensive count values ​​are sorted in descending order, and the first plurality of the underlying index categories with larger comprehensive count values ​​are selected as the interest features.

[0031] Optionally, performing statistics on the underlying index classification to determine the concentration of the underlying index classification includes:

[0032] The concentration is calculated according to formula (1):

[0033]

[0034] Where C is the concentration, N is the number of underlying index categories in the historical query records, and x i is the count value of the ith underlying index category, is the average count value.

[0035] Optionally, each of the count values ​​is updated respectively according to the distance from the current time to obtain a comprehensive count value of the underlying index classification, including:

[0036] The comprehensive count value is updated according to formula (2):

[0037]

[0038] Among them, x i ′ is the comprehensive count value, l is the distance, x il is the count value of distance l in the i-th underlying index category.

[0039] Optionally, constructing a first sensitivity tree of the user according to the interest feature includes:

[0040] Performing a clustering operation on each bottom-level index classification according to the upper-level index classification of each bottom-level index classification;

[0041] Determine whether the number of the underlying index classifications included in each category after the clustering operation is greater than or equal to a minimum threshold;

[0042] In the case where it is determined that the number of the underlying index classifications included in each category after the clustering operation is greater than or equal to the minimum threshold, constructing the first sensitive tree according to the corresponding relationship of each category after the clustering operation;

[0043] When it is determined that the number of underlying index classifications included in any category after the clustering operation is greater than or equal to the minimum threshold, the clustering basis of the clustering operation is corrected to the upper-level index classification of the current clustering basis, the clustering operation is performed again, and the step of determining whether the number of underlying index classifications included in each category after the clustering operation is greater than or equal to the minimum threshold is returned.

[0044] Optionally, obtaining the identity feature of the user includes:

[0045] Obtaining the current identity information of the user;

[0046] Based on the identity information, obtaining in the database a common access index classification corresponding to the identity information;

[0047] The frequently used access index classification is used as the identity feature.

[0048] Optionally, fusing the first sensitivity tree and the second sensitivity tree to obtain a comprehensive sensitivity tree for the user includes:

[0049] Determine a union of the first sensitive tree and the second sensitive tree;

[0050] Starting from the top layers of the first sensitive tree and the second sensitive tree, determining whether each valid index classification of the first sensitive tree and the second sensitive tree is the same;

[0051] When it is determined that the valid index categories are the same, the weight of the corresponding index category in the union is increased by 1 to update the union;

[0052] The union is used as the comprehensive sensitivity tree.

[0053] Optionally, sorting the recall results according to the comprehensive sensitivity tree includes:

[0054] Traversing the documents corresponding to each of the recall results, and searching for corresponding positions in the comprehensive sensitive tree respectively;

[0055] In the case where the document can be found, determining whether the found location is a valid index classification;

[0056] In the case where the found position is a valid index classification, the weight of the corresponding index classification is used as the sorting weight of the recall result;

[0057] In the case where the found position is an invalid index classification, the upper index classification of the found position is used as a tag of the document, and the document is added to the list to be sorted;

[0058] The ranking weights of all documents in the to-be-ranked list are set to the minimum ranking weights of documents in the valid index classification;

[0059] Updating the ranking weight of each of the documents in the to-be-ranked list according to the marks of all the documents therein;

[0060] In the case that the document cannot be found, determining whether the document is one;

[0061] When there is only one document, the ranking weight of the document is set to the minimum ranking weight of the documents that can be found;

[0062] In the case where there are multiple documents, the maximum value of the ranking weight of the document is set as the minimum value of the ranking weight of the document that can be found, and the remaining documents are updated according to the keyword association degree of the document that cannot be found;

[0063] Each of the documents is ranked according to the ranking weight.

[0064] On the other hand, the present invention also provides a semantic retrieval-based multi-way recall ranking analysis system, the system comprising a processor, and the processor is configured to execute any of the methods described above.

[0065] Through the above technical scheme, the embodiment of the present invention provides a sorting and analysis method and system for multi-way recall based on semantic retrieval. The method and system sort the results of multi-way recall by combining the user's interest characteristics and identity characteristics within the system, thereby achieving an improvement in the retrieval method based on multi-way recall and improving the retrieval accuracy and efficiency within the power system.

[0066] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0068] Figure 1 is a flow chart of a multi-way recall ranking analysis method based on semantic retrieval according to an embodiment of the present invention;

[0069] Figure 2 is a flowchart of a method for obtaining a recall result according to an embodiment of the present invention;

[0070] Figure 3 is a flow chart of a method for acquiring interest features according to an embodiment of the present invention;

[0071] Figure 4 is a flowchart of a method for obtaining a first sensitive tree according to an embodiment of the present invention;

[0072] Figure 5 is an example diagram of a first sensitive tree and a second sensitive tree according to an embodiment of the present invention;

[0073] Figure 6 The flowchart of sorting recall results according to a comprehensive sensitivity tree according to one embodiment of the present invention. DETAILED DESCRIPTION

[0074] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0075] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of national laws and regulations. In the embodiments of this application, some existing solutions in the industry such as certain software, components, and models may be mentioned, which should be considered as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0076] like Figure 1 FIG. 1 is a flowchart of a multi-way recall ranking analysis method based on semantic retrieval according to an embodiment of the present invention. Figure 1 In the method, the method may include the following steps:

[0077] In step S10, the keyword input by the user is obtained;

[0078] In step S11, the recall result of each route is obtained according to the keyword;

[0079] In step S12, the user's interest characteristics are obtained;

[0080] In step S13, a first sensitivity tree of the user is constructed according to the interest characteristics;

[0081] In step S14, the identity characteristics of the user are obtained;

[0082] In step S15, a second sensitive tree corresponding to the identity feature is obtained;

[0083] In step S16, the first sensitivity tree and the second sensitivity tree are merged to obtain a comprehensive sensitivity tree of the user;

[0084] In step S17, the recall results are sorted according to the comprehensive sensitivity tree.

[0085] In this Figure 1 In the method shown, step S10 can obtain a keyword input by the user. The keyword can be a single keyword directly input by the user in the search component or extracted from a keyword sentence containing the keyword. Further, in the case where the information input by the user is a keyword sentence containing the keyword, step S10 can also be to first perform semantic analysis on the keyword sentence and determine the keyword in the keyword sentence based on context relevance.

[0086] Step S11 may be used to obtain the recall result of each route according to the keyword. The method for obtaining the recall result of each route may be in various forms known to those skilled in the art. In one example of the present invention, the method for obtaining the recall result may include: Figure 2 The method shown. Figure 2 The method for obtaining the recall result may include the following steps:

[0087] In step S20, an expansion operation is performed on the keyword to obtain a plurality of synonyms;

[0088] In step S21, matching searches are performed in the database based on the keywords and synonyms to obtain recall results;

[0089] In step S22, the keywords and synonyms are encoded according to a preset word library to obtain corresponding keyword codes;

[0090] In step S23, the keyword encoding is input into a preset language model to obtain a corresponding vector representation;

[0091] In step S24, a similarity query is performed in the database according to the vector representation to obtain a recall result.

[0092] Step S12 may be used to obtain the user's interest feature. The interest feature may be used to represent the index category that the user is interested in. The specific method for obtaining the interest feature may be in various forms known to those skilled in the art, including but not limited to calling the user's favorite records, historical access records, etc. In one example of the present invention, the method for obtaining the interest feature may include: Figure 3 In the steps shown in Figure 3 In the method for obtaining the interest feature, the method may include the following steps:

[0093] In step S30, the user's historical query records are obtained;

[0094] In step S31, the historical query records are classified and counted to determine the count value of each underlying index classification corresponding to the historical query records;

[0095] In step S32, statistics are collected on the underlying index classification to determine the concentration of the underlying index classification;

[0096] In step S33, it is determined whether the concentration is greater than or equal to a preset concentration threshold;

[0097] In step S34, when it is determined that the concentration is greater than or equal to the concentration threshold, the underlying index classification whose count value is greater than the preset count value threshold is taken as the interest feature;

[0098] In step S35, when it is determined that the concentration is less than the concentration threshold, the historical query record is divided into a plurality of sub-records according to a preset time interval;

[0099] In step S36, each sub-record is traversed to determine the count value of the underlying index classification included in each sub-record;

[0100] In step S37, each count value is updated according to the distance from the current time to obtain a comprehensive count value of the underlying index classification;

[0101] In step S38, the comprehensive count values ​​are sorted in descending order, and the first multiple underlying index categories with larger comprehensive count values ​​are selected as the determined interest features.

[0102] In this Figure 3 In the method shown, step S30 can be used to obtain the user's historical query record, which can include multiple query records, each of which is in the form of the last link visited by the user after entering a single keyword. Therefore, in an example of the present invention, when obtaining the user's historical query record, step S30 cannot directly crawl all the user's historical access records in the server, but instead selects the last link visited as a query record in the historical query record for each search after the user enters a keyword.

[0103] Step S31 can be used to classify and count the historical query records to determine the count value of each underlying index classification corresponding to the historical query records. Specifically, in this embodiment, step S31 can be to count the underlying index classification according to the corresponding position in the index for the document corresponding to the link of each query record.

[0104] Step S32 can be used to perform statistics on the underlying index classification to determine the concentration of the underlying index classification. The concentration can be calculated in various forms known to those skilled in the art. In one example of the present invention, considering that the concentration is a parameter used to measure the discreteness of the count value of the underlying index classification, the concentration can be calculated using the following formula (1):

[0105]

[0106] Among them, C is the concentration, N is the number of underlying index categories in the historical query records, and x i is the count value of the ith underlying index category, is the average count value.

[0107] Step S33 can be used to determine whether the concentration is greater than or equal to a preset concentration threshold. If the concentration is greater than the concentration threshold, it means that the user has been accessing some underlying index categories in a concentrated manner at any time, so the underlying index categories with count values ​​greater than or equal to the count value threshold can be directly used as interest features, that is, step S34.

[0108] On the contrary, it means that the user's access is relatively scattered, and since the relatively scattered underlying index classification is not conducive to subsequent sorting and screening, it is necessary to further process based on the time series through steps 35 to S38. Specifically, in this embodiment, in step S35, when it is judged that the concentration is less than the concentration threshold, the historical query record is divided into multiple sub-records according to the preset time interval, and step S36 is used to traverse each sub-record and determine the count value of the underlying index classification included in each sub-record, so as to determine the user's access habits in different time periods. Considering that the farther the user's sub-record is from the current time, the less it can reflect the user's access habits, it is not possible to directly represent the desired comprehensive count value by simply adding the count values ​​of each sub-record. In one example of the present invention, considering the influence of time distance, the comprehensive count value can be calculated using the following formula (2):

[0109]

[0110] Among them, x i ′ is the comprehensive count value, l is the distance, x il is the count value of distance l in the i-th underlying index category.

[0111] Step S38 is used to sort in descending order according to the comprehensive count value, and select the first multiple underlying index categories with larger comprehensive count values ​​as the determined interest features. Among them, the method for determining the number of selected underlying index categories can be a variety of forms known to those skilled in the art, such as by presetting the number. In one example of the present invention, since it is necessary to select the most representative ones and it is not possible to select all the underlying index categories, the difference in count values ​​of adjacent underlying index categories after sorting is predetermined, and then the screening is performed in order from left to right. When the difference is greater than a preset threshold, the selection can be stopped, the underlying index categories after the difference are discarded, and the underlying index categories before the difference are used as the interest features.

[0112] Step S13 is used to construct a first sensitive tree of the user according to the interest feature. The first sensitive tree can be used to represent the distribution characteristics of the index classification of the user based on the interest feature. The specific method for obtaining the first sensitive tree can be various forms known to those skilled in the art. In one example of the present invention, the method for obtaining the first sensitive tree may include: Figure 4 In the steps shown in Figure 4 In the method for obtaining the first sensitive tree, the method may include the following steps:

[0113] In step S40, a clustering operation is performed on each bottom-level index classification according to the upper-level index classification of each bottom-level index classification;

[0114] In step S41, it is determined whether the number of underlying index categories included in each category after the clustering operation is greater than or equal to a minimum threshold;

[0115] In step S42, when it is determined that the number of underlying index categories included in each category after the clustering operation is greater than or equal to the minimum threshold, a first sensitive tree is constructed according to the corresponding relationship of each category after the clustering operation;

[0116] In step S43, when it is determined that the number of underlying index classifications included in any category after the clustering operation is greater than or equal to the minimum threshold, the clustering basis of the clustering operation is corrected to the upper-level index classification of the current clustering basis, the clustering operation is performed again, and the step of determining whether the number of underlying index classifications included in each category after the clustering operation is greater than or equal to the minimum threshold is returned, that is, step S41 is returned.

[0117] Step S14 can be used to obtain the user's identity features. The identity features can be used to represent the user's access habits to the identity information in the current system. Unlike interest features, since there are many users with the same identity information in the same system, the number of samples that can be provided is also large. At this time, it is only necessary to directly obtain the commonly used access index classification of the user's identity information as the identity feature.

[0118] Step S15 may be used to obtain a second sensitive tree corresponding to the identity feature. The second sensitive tree may be obtained in a manner similar to the first sensitive tree, and the manner of obtaining the first sensitive tree has been described in detail above, so it will not be repeated here.

[0119] Step S16 may be used to merge the first sensitive tree and the second sensitive tree to obtain a comprehensive sensitive tree for the user. Specifically, in one example of the present invention, an example diagram of the first sensitive tree and the second sensitive tree may be as follows: Figure 5 As shown. Figure 5In the figure, the first sensitive tree and the second sensitive tree both include multiple index categories, wherein the index category number suffixed with "*" indicates a valid index category, which indicates that the user is interested in the index category; otherwise, it is an invalid index category, which indicates that the user may be interested in the upper-level index category of the index category.

[0120] The step S16 may be to first compare the bottom index classification of each first sensitive tree and the second sensitive tree respectively, determine whether there is the same bottom index classification, and then, if it is determined that there is the same bottom index classification, add 1 to the weight of the corresponding upper index classification. Based on this example, in an example of the present invention, the method for obtaining the comprehensive sensitive tree may be to first determine the union of the first sensitive tree and the second sensitive tree, and then, starting from the top layers of the first sensitive tree and the second sensitive tree, respectively, determine whether each valid index classification of the first sensitive tree and the second sensitive tree is the same, and if it is determined that the valid index classification is the same, increase the weight of the corresponding index classification in the union by 1 to update the union, and finally output the union as a comprehensive sensitive tree.

[0121] Finally, step S17 may be used to sort the recall results according to the comprehensive sensitivity tree. Specifically, in one example of the present invention, the method for sorting the recall results according to the comprehensive sensitivity tree may include: Figure 6 In the steps shown in Figure 6 In the method for sorting the recall results according to the comprehensive sensitivity tree, the method may include the following steps:

[0122] In step S50, the documents corresponding to each recall result are traversed, and the corresponding positions are searched in the comprehensive sensitivity tree respectively;

[0123] In step S51, when the document can be found, it is determined whether the found location is a valid index classification;

[0124] In step S52, when the found position is a valid index classification, the weight of the corresponding index classification is used as the ranking weight of the recall result;

[0125] In step S53, when the found position is an invalid index classification, the upper index classification of the found position is used as a tag of the document, and the document is added to the list to be sorted;

[0126] In step S54, the ranking weights of all documents in the to-be-ranked list are set to the minimum ranking weights of documents in the valid index classification;

[0127] In step S55, the ranking weight of each document in the to-be-ranked list is updated according to the tags of all the documents in the to-be-ranked list. Specifically, in this example, the method for updating the ranking weight may be to use the following formula (3) for updating:

[0128]

[0129] Among them, λ i ′ is the updated ranking weight, λ i is the ranking weight before updating, c i is the tag of the i-th document, and s is the number of documents in the list to be sorted.

[0130] In step S56, if the document cannot be found, it is determined whether the document is one;

[0131] In step S57, when there is only one document, the ranking weight of the document is set to the minimum ranking weight of the documents that can be found;

[0132] In step S58, when there are multiple documents, the maximum value of the ranking weight of the document is set to the minimum value of the ranking weight of the document that can be found, and the remaining documents are updated according to the keyword association degree of the document that cannot be found. Specifically, the keyword association degree can be the sorting order based on the keyword query in the recall result. In order to ensure that the documents in this part can meet the order during sorting, the ranking weight of the documents therein can be set to start from the maximum value and decrease by a fixed value in sequence. The fixed value can be 0.1 times the maximum value.

[0133] In step S59, each document is sorted according to the sorting weight.

[0134] On the other hand, the present invention also provides a semantic retrieval-based multi-way recall ranking analysis system, the system comprising a processor, and the processor is configured to execute any of the methods described above.

[0135] Through the above technical scheme, the embodiment of the present invention provides a sorting and analysis method and system for multi-way recall based on semantic retrieval. The method and system sort the results of multi-way recall by combining the user's interest characteristics and identity characteristics within the system, thereby achieving an improvement in the retrieval method based on multi-way recall and improving the retrieval accuracy and efficiency within the power system.

[0136] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0137] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0140] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0141] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0142] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0143] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0144] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A multi-way recall ranking analysis method based on semantic retrieval, characterized in that: The method comprises: Get the keywords entered by the user; Obtaining the recall result of each route according to the keyword; Acquire interest characteristics of the user; Constructing a first sensitivity tree of the user according to the interest feature; Obtaining identity characteristics of the user; Obtaining a second sensitive tree corresponding to the identity feature; Merging the first sensitivity tree and the second sensitivity tree to obtain a comprehensive sensitivity tree for the user; The recall results are sorted according to the comprehensive sensitivity tree.

2. The method according to claim 1, characterized in that The recall results of each channel are obtained according to the keywords, including: Performing an expansion operation on the keyword to obtain multiple synonyms; Perform matching search in a database according to the keywords and synonyms to obtain the recall result; Encoding the keywords and synonyms according to a preset word library to obtain corresponding keyword codes; Inputting the keyword encoding into a preset language model to obtain a corresponding vector representation; A similarity query is performed in the database according to the vector representation to obtain the recall result.

3. The method according to claim 1, characterized in that Obtaining the user's interest characteristics includes: Get the user's historical query records; Classify and count the historical query records to determine the count value of each underlying index classification corresponding to the historical query records; Performing statistics on the underlying index classification to determine the concentration of the underlying index classification; Determining whether the concentration is greater than or equal to a preset concentration threshold; In the case where it is determined that the concentration is greater than or equal to the concentration threshold, classifying the underlying index having a count value greater than a preset count value threshold as the interest feature; In the case where it is determined that the concentration is less than the concentration threshold, dividing the historical query record into a plurality of sub-records according to a preset time interval; Traversing each of the sub-records to determine the count value of the underlying index classification included in each of the sub-records; updating each of the count values ​​according to the distance from the current time to obtain a comprehensive count value of the underlying index classification; The comprehensive count values ​​are sorted in descending order, and the first plurality of the underlying index categories with larger comprehensive count values ​​are selected as the interest features.

4. The method according to claim 3, characterized in that Performing statistics on the underlying index classification to determine the concentration of the underlying index classification includes: The concentration is calculated according to formula (1): Where C is the concentration, N is the number of underlying index categories in the historical query records, and x i is the count value of the ith underlying index category, is the average count value.

5. The method according to claim 3, characterized in that: Each of the count values ​​is updated according to the distance from the current time to obtain a comprehensive count value of the underlying index classification, including: The comprehensive count value is updated according to formula (2): Among them, x i ′ is the comprehensive count value, l is the distance, x il is the count value of distance l in the i-th underlying index category.

6. The method according to claim 1, characterized in that Constructing a first sensitivity tree of the user according to the interest feature includes: Performing a clustering operation on each bottom-level index classification according to the upper-level index classification of each bottom-level index classification; Determine whether the number of the underlying index classifications included in each category after the clustering operation is greater than or equal to a minimum threshold; In the case where it is determined that the number of the underlying index classifications included in each category after the clustering operation is greater than or equal to the minimum threshold, constructing the first sensitive tree according to the corresponding relationship of each category after the clustering operation; When it is determined that the number of underlying index classifications included in any category after the clustering operation is greater than or equal to the minimum threshold, the clustering basis of the clustering operation is corrected to the upper-level index classification of the current clustering basis, the clustering operation is performed again, and the step of determining whether the number of underlying index classifications included in each category after the clustering operation is greater than or equal to the minimum threshold is returned.

7. The method according to claim 1, characterized in that Obtaining the identity characteristics of the user, including: Obtaining the current identity information of the user; Based on the identity information, obtaining in a database a commonly used access index classification corresponding to the identity information; The frequently used access index classification is used as the identity feature.

8. The method according to claim 1, characterized in that The first sensitivity tree and the second sensitivity tree are merged to obtain a comprehensive sensitivity tree of the user, including: Determine a union of the first sensitive tree and the second sensitive tree; Starting from the top layers of the first sensitive tree and the second sensitive tree, determining whether each valid index classification of the first sensitive tree and the second sensitive tree is the same; When it is determined that the valid index categories are the same, the weight of the corresponding index category in the union is increased by 1 to update the union; The union is used as the comprehensive sensitivity tree.

9. The method according to claim 8, characterized in that Sorting the recall results according to the comprehensive sensitivity tree includes: Traversing the documents corresponding to each of the recall results, and searching for corresponding positions in the comprehensive sensitive tree respectively; In the case where the document can be found, determining whether the found location is a valid index classification; In the case where the found position is a valid index classification, the weight of the corresponding index classification is used as the sorting weight of the recall result; In the case where the found position is an invalid index classification, the upper index classification of the found position is used as a tag of the document, and the document is added to the list to be sorted; The ranking weights of all documents in the to-be-ranked list are set to the minimum ranking weights of documents in the valid index classification; Updating the ranking weight of each of the documents in the to-be-ranked list according to the marks of all the documents therein; In the case that the document cannot be found, determining whether the document is one; When there is only one document, the ranking weight of the document is set to the minimum ranking weight of the documents that can be found; In the case where there are multiple documents, the maximum value of the ranking weight of the document is set as the minimum value of the ranking weight of the document that can be found, and the remaining documents are updated according to the keyword association degree of the document that cannot be found; Each of the documents is ranked according to the ranking weight.

10. A multi-way recall ranking analysis system based on semantic retrieval, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 9.

Images