CN113254810B

CN113254810B - Search result output method and device, computer equipment and readable storage medium

Info

Publication number: CN113254810B
Application number: CN202110673372.0A
Authority: CN
Inventors: 苑爱泉
Original assignee: Zhejiang Koubei Network Technology Co Ltd
Current assignee: Zhejiang Koubei Network Technology Co Ltd
Priority date: 2021-06-17
Filing date: 2021-06-17
Publication date: 2021-10-29
Anticipated expiration: 2041-06-17
Also published as: CN113254810A; WO2022262849A1

Abstract

The application discloses a search result output method, a search result output device, a computer device and a readable storage medium, relates to the technical field of internet, considers the relevance between scenes and search results in a sorting process, realizes diversified sorting, ensures that the search results arranged in the front are attached to the search requirements of users, improves the accuracy of the output search results, and is good in universality. The method comprises the following steps: acquiring an initial result queue; dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result in the plurality of search results and the search content and the geographic position of the user; sorting the search results included in each of the plurality of information groups according to a plurality of preset sorting factors to obtain an intermediate result queue; adjusting the sequence of the search results included in the intermediate result queue based on the relevance between the search results of the same order in the initial result queue and the intermediate result queue to obtain a target result queue; and outputting the target result queue.

Description

Search result output method and device, computer equipment and readable storage medium

Technical Field

The present application relates to the field of internet technologies, and in particular, to a method and an apparatus for outputting search results, a computer device, and a readable storage medium.

Background

In recent years, with the rapid development of internet technology, various internet applications are widely deepened into various fields, big data is explosively increased, massive data and information are dispersed in a network space, and when a user needs to acquire the information and the data, information search can be performed through a search platform, so that the search platform can output related search contents. The search platform, as an important link between the user and the information, generally outputs a large number of related search results according to the content searched by the user, and provides enough information for the user as possible for reference.

In the related technology, a search platform queries a series of results related to search content according to the search content input by a user, performs CTR (Click-Through-Rate) estimation on the results, then evaluates the estimated results according to a preset standard, further sorts the results, and outputs the sorted results to the user for viewing.

In carrying out the present application, the applicant has found that the related art has at least the following problems:

the search results output after the information is sorted based on the preset standard have limitations, and the search results arranged in the front may not meet the real search requirements of the user, so that the output search results are not accurate enough, and the universality is poor.

Disclosure of Invention

In view of this, the present application provides a method and an apparatus for outputting search results, a computer device and a readable storage medium, and mainly aims to solve the problems that the current search results have limitations, and the search results ranked in the front may not meet the real search requirements of the user, which results in inaccurate output search results and poor versatility.

According to a first aspect of the present application, there is provided a search result output method, the method including:

obtaining an initial result queue, wherein the initial result queue comprises a plurality of search results which are determined based on search contents input by a user and are related to the search contents;

dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result in the plurality of search results and the search content and the geographic position of the user;

sorting the search results included in each of the plurality of information groups according to a plurality of preset sorting factors to obtain an intermediate result queue;

based on the relevance between the initial result queue and the search result of the same order in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain a target result queue;

and outputting the target result queue.

Optionally, the obtaining an initial result queue includes:

receiving the search content input by the user;

analyzing the search content, and inquiring the plurality of search results related to the search content;

performing Click Through Rate (CTR) estimation on the plurality of search results, and outputting a plurality of estimated click rates of the plurality of search results;

and sequencing the plurality of search results according to the plurality of estimated click rates to obtain the initial result queue.

Optionally, the dividing the plurality of search results into a plurality of information groups according to a matching degree between each search result of the plurality of search results and the search content and the search scene in which the user searches includes:

for each search result in the plurality of search results, determining a semantic matching degree and a scene matching degree of each search result, wherein the semantic matching degree indicates the semantic relevance degree of the search result and the search content, and the scene matching degree indicates the relevance degree of the search result and the search scene;

determining a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals;

comparing the semantic matching degree and the scene matching degree of each search result with the semantic matching degree value interval and the scene matching degree value intervals, and dividing the search results with the semantic matching degree and the scene matching degree in the same value interval into the same group to obtain a plurality of information groups.

Optionally, the determining the semantic matching degree and the scene matching degree of each search result includes:

for each search result in the plurality of search results, querying the semantic matching degree between the search result and the search content;

determining the geographic position of the user when the user inputs the search content and the store positions of stores related to the search result;

and calculating the position distance between the geographic position and the store position, and acquiring the scene matching degree indicated by the position distance.

Optionally, the determining a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals includes:

inquiring a preset division standard, and extracting the plurality of semantic matching degree value intervals and the plurality of scene matching degree value intervals from the preset division standard; or the like, or, alternatively,

counting a first preset number of sample parameters, performing semantic training on the first preset number of sample parameters based on a machine learning depth model to obtain a plurality of semantic matching degree values, constructing a plurality of semantic matching degree value intervals according to the size relationship among the plurality of semantic matching degree values, performing scene training on the first preset number of sample parameters based on the machine learning depth model to obtain a plurality of scene matching degree values, and constructing the plurality of scene matching degree value intervals according to the size relationship among the plurality of scene matching degree values.

Optionally, the sorting the search results included in each of the plurality of information groups according to a plurality of preset sorting factors to obtain an intermediate result queue includes:

for each information group in the plurality of information groups, calculating a relevance score of each search result included in each information group based on the plurality of preset ranking factors;

sequencing all search results included in the information group according to the sequence of the relevance scores from large to small to obtain the sequenced information group;

calculating and sequencing relevance scores for search results included in each of the plurality of information groups respectively to obtain a plurality of sequenced information groups;

and combining the sequenced information groups according to the sequence of the grouping grades corresponding to the information groups from high to low to obtain the intermediate result queue.

Optionally, the calculating a relevance score of each search result included in each information group based on the plurality of preset ranking factors includes:

for each of the search results included in the set of information, querying a plurality of factor scores of the search result over the plurality of preset ranking factors;

acquiring a plurality of factor weights corresponding to the preset ranking factors, and performing weight calculation on the factor scores based on the factor weights to obtain the relevance scores of the search results;

and respectively calculating the relevance score for each search result to obtain the relevance score of each search result.

Optionally, the obtaining of the multiple factor weights corresponding to the multiple preset ranking factors includes:

for each preset ranking factor in the plurality of preset ranking factors, inquiring the search intention and the search industry corresponding to the search content input by the user;

and training the search intention, the search industry and each preset ranking factor respectively based on a linear function to obtain the multiple factor weights.

Optionally, the adjusting, based on the correlation between the initial result queue and the search result of the same rank in the intermediate result queue, the order of the search results included in the intermediate result queue to obtain a target result queue includes:

comparing every two search results in the initial result queue and the intermediate result queue at the same time to determine the relevance between the two search results;

determining the sequence of the two search results and adjusting the intermediate result queue according to the relevance between the two search results to obtain the adjusted intermediate result queue;

inputting the adjusted intermediate result queue to an evaluator for evaluation to obtain a queue score output by the evaluator, and labeling the adjusted intermediate result queue by using the queue score, wherein the evaluator is trained based on a plurality of sample queues and indicates the score of each sample queue in the plurality of sample queues;

repeatedly executing the comparison process, comparing the adjusted intermediate result queue with the initial result queue and readjusting the adjusted intermediate result queue until the adjustment times reach the preset rotation times, and obtaining queue scores of which the number meets the preset rotation times;

extracting a target queue score with the highest queue score and a target intermediate result queue labeled by the target queue score from the queue scores with the number meeting the preset cycle times;

and determining a second preset number, intercepting the search results of which the number meets the second preset number at the head of the target intermediate result queue, and taking the search results of which the number meets the second preset number as the target result queue.

Optionally, after the adjusting the order of the search results included in the intermediate result queue based on the correlation between the initial result queue and the search result of the same rank in the intermediate result queue to obtain the target result queue, the method further includes:

inquiring a preset regulation and control strategy, and determining the regulation and control requirement of the preset regulation and control strategy;

sequentially adjusting the search results included in the target result queue according to the regulation and control requirement to obtain the regulated and controlled target result queue;

and outputting the regulated and controlled target result queue.

According to a second aspect of the present application, there is provided a search result output apparatus including:

an obtaining module, configured to obtain an initial result queue, where the initial result queue includes a plurality of search results related to search content determined based on the search content input by a user;

the dividing module is used for dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result in the plurality of search results, the search content and the geographic position of the user;

the sorting module is used for sorting the search results included by each information group in the plurality of information groups according to a plurality of preset sorting factors to obtain an intermediate result queue;

the adjusting module is used for adjusting the sequence of the search results contained in the intermediate result queue to obtain a target result queue based on the relevance between the initial result queue and the search result of the same rank in the intermediate result queue;

and the output module is used for outputting the target result queue.

Optionally, the obtaining module is configured to receive the search content input by the user; analyzing the search content, and inquiring the plurality of search results related to the search content; performing Click Through Rate (CTR) estimation on the plurality of search results, and outputting a plurality of estimated click rates of the plurality of search results; and sequencing the plurality of search results according to the plurality of estimated click rates to obtain the initial result queue.

Optionally, the dividing module is configured to determine, for each search result in the plurality of search results, a semantic matching degree and a scene matching degree of the each search result, where the semantic matching degree indicates a semantic relevance degree of the search result to the search content, and the scene matching degree indicates a relevance degree of the search result to the search scene; determining a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals; comparing the semantic matching degree and the scene matching degree of each search result with the semantic matching degree value interval and the scene matching degree value intervals, and dividing the search results with the semantic matching degree and the scene matching degree in the same value interval into the same group to obtain a plurality of information groups.

Optionally, the dividing module is configured to, for each search result in the plurality of search results, query the semantic matching degree between the search result and the search content; determining the geographic position of the user when the user inputs the search content and the store positions of stores related to the search result; and calculating the position distance between the geographic position and the store position, and acquiring the scene matching degree indicated by the position distance.

Optionally, the dividing module is configured to query a preset dividing standard, and extract the multiple semantic matching degree value intervals and the multiple scene matching degree value intervals from the preset dividing standard; or counting a first preset number of sample parameters, performing semantic training on the first preset number of sample parameters based on a machine learning depth model to obtain a plurality of semantic matching degree values, constructing a plurality of semantic matching degree value intervals according to the size relationship among the plurality of semantic matching degree values, performing scene training on the first preset number of sample parameters based on the machine learning depth model to obtain a plurality of scene matching degree values, and constructing the plurality of scene matching degree value intervals according to the size relationship among the plurality of scene matching degree values.

Optionally, the ranking module is configured to calculate, for each information group in the plurality of information groups, a relevance score of each search result included in each information group based on the plurality of preset ranking factors; sequencing all search results included in the information group according to the sequence of the relevance scores from large to small to obtain the sequenced information group; calculating and sequencing relevance scores for search results included in each of the plurality of information groups respectively to obtain a plurality of sequenced information groups; and combining the sequenced information groups according to the sequence of the grouping grades corresponding to the information groups from high to low to obtain the intermediate result queue.

Optionally, the ranking module is configured to, for each search result included in the information group, query a plurality of factor scores of the search result over the plurality of preset ranking factors; acquiring a plurality of factor weights corresponding to the preset ranking factors, and performing weight calculation on the factor scores based on the factor weights to obtain the relevance scores of the search results; and respectively calculating the relevance score for each search result to obtain the relevance score of each search result.

Optionally, the ranking module is configured to query, for each preset ranking factor in the plurality of preset ranking factors, a search intention and a search industry corresponding to search content input by a user; and training the search intention, the search industry and each preset ranking factor respectively based on a linear function to obtain the multiple factor weights.

Optionally, the adjusting module is configured to compare every two search results in the initial result queue and the intermediate result queue at the same rank, and determine a relevance between the two search results; determining the sequence of the two search results and adjusting the intermediate result queue according to the relevance between the two search results to obtain the adjusted intermediate result queue; inputting the adjusted intermediate result queue to an evaluator for evaluation to obtain a queue score output by the evaluator, and labeling the adjusted intermediate result queue by using the queue score, wherein the evaluator is trained based on a plurality of sample queues and indicates the score of each sample queue in the plurality of sample queues; repeatedly executing the comparison process, comparing the adjusted intermediate result queue with the initial result queue and readjusting the adjusted intermediate result queue until the adjustment times reach the preset rotation times, and obtaining queue scores of which the number meets the preset rotation times; extracting a target queue score with the highest queue score and a target intermediate result queue labeled by the target queue score from the queue scores with the number meeting the preset cycle times; and determining a second preset number, intercepting the search results of which the number meets the second preset number at the head of the target intermediate result queue, and taking the search results of which the number meets the second preset number as the target result queue.

Optionally, the apparatus further comprises:

the query module is used for querying a preset regulation and control strategy and determining the regulation and control requirement of the preset regulation and control strategy;

the adjusting module is further configured to sequentially adjust the search results included in the target result queue according to the adjustment and control requirement, so as to obtain the adjusted and controlled target result queue;

and the output module is also used for outputting the regulated and controlled target result queue.

According to a third aspect of the present application, there is provided a computer device comprising a memory storing a computer program and a processor implementing the steps of the method of any of the first aspects when the computer program is executed.

According to a fourth aspect of the present application, there is provided a readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of the method of any of the above-mentioned first aspects.

By means of the technical scheme, according to the search result output method, the search result output device, the computer equipment and the readable storage medium, after an initial result queue is obtained according to search contents of a user, the search results in the initial result queue are divided into a plurality of information groups according to the matching degree between each search result in the initial result queue and the search contents and the geographic position of the user, the search results included in the information groups are sorted in groups according to a plurality of preset sorting factors, and the sorted information groups are integrated to obtain an intermediate result queue. And then, based on the relevance between the initial result queue and the search result of the same rank in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain and output a target result queue, so that the scene where the user searches and the relevance between the search results are considered in the sorting process of the target result queue, the diversified sorting of the search results is realized, the search results arranged in the front in the target result queue are ensured to be attached to the search requirements of the user, the accuracy of the output search results is improved, and the universality is better.

The foregoing description is only an overview of the technical solutions of the present application, and the present application can be implemented according to the content of the description in order to make the technical means of the present application more clearly understood, and the following detailed description of the present application is given in order to make the above and other objects, features, and advantages of the present application more clearly understandable.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the application. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

fig. 1 is a schematic flowchart illustrating a search result output method according to an embodiment of the present application;

fig. 2A is a schematic flowchart illustrating a search result output method according to an embodiment of the present application;

FIG. 2B is a diagram illustrating a search result output method according to an embodiment of the present disclosure;

fig. 3A is a schematic structural diagram illustrating a search result output apparatus according to an embodiment of the present application;

fig. 3B is a schematic structural diagram illustrating a search result output device according to an embodiment of the present application;

fig. 4 shows a schematic device structure diagram of a computer apparatus according to an embodiment of the present application.

Detailed Description

Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

An embodiment of the present application provides a search result output method, as shown in fig. 1, the method includes:

101. an initial result queue is obtained, the initial result queue including a plurality of search results related to search content determined based on the search content input by a user.

102. And dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result in the plurality of search results, the search content and the geographic position of the user.

103. And sequencing the search results included in each of the plurality of information groups according to the plurality of preset sequencing factors to obtain an intermediate result queue.

104. And adjusting the sequence of the search results included in the intermediate result queue based on the relevance between the search results of the same order in the initial result queue and the intermediate result queue to obtain a target result queue.

105. And outputting the target result queue.

According to the method provided by the embodiment of the application, after the initial result queue is obtained according to the search content of the user, the search results in the initial result queue are divided into a plurality of information groups according to the matching degree between each search result in the initial result queue and the search content and the geographic position of the user, the search results included in the information groups are sorted in groups according to a plurality of preset sorting factors, and the sorted information groups are integrated to obtain the intermediate result queue. And then, based on the relevance between the initial result queue and the search result of the same rank in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain and output a target result queue, so that the scene where the user searches and the relevance between the search results are considered in the sorting process of the target result queue, the diversified sorting of the search results is realized, the search results arranged in the front in the target result queue are ensured to be attached to the search requirements of the user, the accuracy of the output search results is improved, and the universality is better.

An embodiment of the present application provides a search information output method, as shown in fig. 2A, the method includes:

201. an initial result queue is obtained.

The search platform can provide the largest entrance for information search for users, is an important link for connecting users and information, and has become an essential part of people's life as the most common search tool. When the search platform provides search service for users, a large number of related search results can be searched according to search content input by the users, and the search results are considered to have different advantages and disadvantages, so a sequencing mechanism is designed in the search platform, the obtained search results can be sequenced according to the sequencing mechanism, and the sequenced search results are output to the users. Currently, when many search platforms rank search results, a model such as CTR (Click Through Rate) or CVR (Conversion Rate) is usually used to estimate an estimated Click Rate of each search result, and search results with a large estimated Click Rate are ranked in front, so that a user can obtain a high-quality search result Through search first. Or some search platforms directly set business rules and sort the search results according to the business rules. However, the applicant recognizes that there are many points to be considered in the actual search result ranking, for example, considering the search experience of the user, the search efficiency, the merchant appeal, and the like, and the ranking of the search results simply depending on the estimated click rate or the business rule is not flexible enough, and although the interpretability is strong, the final ranking result is not smooth and is not an optimal ranking mode. And the sequencing mode abandons complex and high-dimensional service parameters, is difficult to be used universally in different search scenes, and has poor popularization.

Therefore, the application provides a search information output method, after an initial result queue is obtained according to search contents of a user, according to matching degrees between each search result in the initial result queue and the search contents and geographic positions of the user, the search results in the initial result queue are divided into a plurality of information groups, the search results included in the information groups are sorted in groups according to a plurality of preset sorting factors, and the sorted information groups are integrated to obtain an intermediate result queue. And then, based on the relevance between the initial result queue and the search result of the same rank in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain and output a target result queue, so that the scene where the user searches and the relevance between the search results are considered in the sorting process of the target result queue, the diversified sorting of the search results is realized, the search results arranged in the front in the target result queue are ensured to be attached to the search requirements of the user, the accuracy of the output search results is improved, and the universality is better.

In order to implement the technical scheme of the present application, an initial result queue needs to be obtained first, and subsequent grouping, in-group sorting, rearranging, adjusting, and other processes are performed on the basis of the initial result queue. Wherein the initial result queue includes a plurality of search results related to the search content determined based on the search content input by the user. When generating the initial result queue, firstly, receiving search content input by a user, analyzing the search content, and inquiring a plurality of search results related to the search content. And then performing CTR estimation on the plurality of search results, and outputting a plurality of estimated click rates of the plurality of search results. And finally, sequencing the plurality of search results according to the plurality of estimated click rates to obtain an initial result queue.

202. And dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result in the plurality of search results, the search content and the geographic position of the user.

In the embodiment of the present application, after the initial result queue is obtained, the plurality of search results need to be divided into a plurality of information groups according to the matching degree between each search result in the plurality of search results and the search content and the geographic location where the user is located, and experience-based ranking is established to ensure the interpretability of the search results.

The process of dividing the plurality of search results into the plurality of information groups is actually a process of classifying the search results, and in a search scene, search experiences in at least two dimensions need to be considered when the plurality of information groups are divided. The first dimension needs to consider semantic matching degree, wherein the semantic matching degree refers to semantic correlation degree between a search result and search content; the second dimension needs to consider scene matching degree, which refers to the degree of correlation between the search result and the search scene, and may specifically be the matching proximity between the search result and LBS (Location Based Service) of the user, physics, and space time, and more commonly, the distance. In this way, the process of dividing the plurality of search results into a plurality of information groups is specifically as follows:

first, for each search result in a plurality of search results, a semantic matching degree and a scene matching degree of each search result need to be determined. Specifically. For each search result in the plurality of search results, the semantic matching degree between the search result and the search content can be directly inquired, the geographic position where the user inputs the search content and the store position of the store related to the search result are determined, the position distance between the geographic position and the store position is calculated, and the scene matching degree indicated by the position distance is obtained.

And then, determining a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals. The semantic matching degree value interval and the scene matching degree value interval are used for simultaneously integrating parameters of two dimensions to divide information groups. Specifically, 3 semantic matching degree value intervals and 3 scene matching degree value intervals may be set. For example, the 3 semantic matching degree value intervals can be respectively (0-30%), (30-60%) and (60-90%), (0-30%) interval labels are "irrelevant", (30-60%) interval labels are "weak relevant", and (60-90%) interval labels are "strong relevant". The 3 scene matching degree value intervals can be (0-500 m), (500-2000 m) and (more than 2000 m), the interval label of (0-500 m) is 'short distance', the interval label of (500-2000 m) is 'middle distance', and the interval label of (more than 2000 m) is 'long distance'.

It should be noted that, because the threshold values that the search results from different industries can bear are different, for example, when a user searches an entertainment place, the user can bear a longer distance, and the scene matching degree value range corresponding to the "near" range label can be actually expanded to (0-5000 meters). And all search results obtained by searching in some industries are probably within 5000 meters, belong to a short distance interval, cannot distinguish gears, and have no division significance. Therefore, in the embodiment of the present application, a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals may be determined in two ways, one way is that a preset division standard is manually set by a worker according to different industries, and the plurality of semantic matching degree value intervals and the plurality of scene matching degree value intervals are defined in the preset division standard, so that the search platform directly queries the preset division standard, and extracts the plurality of semantic matching degree value intervals and the plurality of scene matching degree value intervals in the preset division standard. The other method is to count a first preset number of sample parameters, for example, 10 ten thousand of sample parameters, perform semantic training on the first preset number of sample parameters based on a machine learning depth model to obtain a plurality of semantic matching degree values, construct a plurality of semantic matching degree value intervals according to the size relationship among the plurality of semantic matching degree values, perform scene training on the first preset number of sample parameters based on the machine learning depth model to obtain a plurality of scene matching degree values, construct a plurality of scene matching degree value intervals according to the size relationship among the plurality of scene matching degree values, form the plurality of value intervals by using an online dynamic clustering algorithm, and ensure that the value intervals fit the search scene of the user.

And finally, comparing the semantic matching degree and the scene matching degree of each search result with a semantic matching degree value interval and a plurality of scene matching degree value intervals, dividing the search results of which the semantic matching degree and the scene matching degree are in the same value interval into the same group, obtaining a plurality of information groups, and realizing the nested sorting of two dimensions. Continuing with the above example of setting 3 semantic matching degree value intervals and 3 scene matching degree value intervals, when dividing the information group, a two-dimensional orthogonal grid as shown in fig. 2B may be actually established according to boolean logic. One dimension in the two-dimensional orthogonal grid is 3 semantic matching degree value intervals which are respectively 'strong correlation', 'weak correlation' and 'irrelevant', and the other dimension is 3 scene matching degree value intervals which are respectively 'near distance', 'intermediate distance' and 'far distance'. In this way, when the information groups are divided, the search results with the semantic matching degree of "strong correlation" and the scene matching degree of "close range" are divided into the information groups corresponding to the shadow region, the search results with the semantic matching degree of "weak correlation" and the scene matching degree of "intermediate range" are divided into the information groups corresponding to the blank region, and the search results with the semantic matching degree of "irrelevant" and the scene matching degree of "far distance" are divided into the information groups corresponding to the black region. It should be noted that in practical applications, the semantic matching degree of some search results is "weak correlation", and the scene matching degree is "close range", in which case, the search results are divided into information groups corresponding to the blank regions according to the dividing manner in fig. 2B. In addition, in order to distinguish a plurality of information groups and facilitate subsequent integration of the information groups, an information group name may be set for each information group. With continued reference to fig. 2B, since the search result included in the information group corresponding to the shaded area is a search result with excellent correlation and close distance, the information group name of the information group corresponding to the shaded area may be "excellent"; the search result included in the information group corresponding to the blank area is a search result with weak correlation and a medium distance, so the information group name of the information group corresponding to the blank area can be "good"; and the search result included in the information group corresponding to the black area is a search result that is irrelevant in correlation and is distant from the user, the information group name of the information group corresponding to the black area may be "poor".

203. And sequencing the search results included in each of the plurality of information groups according to the plurality of preset sequencing factors to obtain an intermediate result queue.

In the embodiment of the present application, after the information groups are divided, the search results included in each of the information groups need to be sorted according to a plurality of preset sorting factors to obtain an intermediate result queue, so as to implement intra-group sorting of the information groups and integration of the information groups. The preset ranking factors may be one or more of a query matching degree, a scene matching degree, a user matching degree, and a shop quality measure. The query matching degree can be text correlation, category consistency, entity consistency, knowledge correlation, general large model score and the like between the search result and the search content; the scene matching degree can be distance grading, distance smooth grading, POI (Point Of interest) type matching and the like Of the search result; the user matching degree can be the estimated click rate and the conversion rate corresponding to the search result; the store quality metric may be a store business quality score, a store material quality score, etc. of the store to which the search result corresponds. Since each search result corresponds to a plurality of different factor scores on a plurality of preset ranking factors, ranking cannot be achieved, therefore, the plurality of factor scores need to be fused into one relevance score, and intra-group ranking is achieved according to the relevance score. The generation process of the relevance score is described below by taking any search result in any information group as an example:

first, a plurality of factor scores of the search results over a plurality of preset ranking factors are queried. Subsequently, obtaining a plurality of factor weights corresponding to a plurality of preset ranking factors, and performing weight calculation on the plurality of factor scores based on the plurality of factor weights to obtain a relevance score of the search result, wherein a formula of the weight calculation is shown as formula 1:

equation 1: correlation score = w1 × score1+ w2 × score2+ ….

Wherein score1 and score2 are two factor scores, and w1 and w2 are factor weights corresponding to the two factor scores. It should be noted that, considering that the sensitivity degrees of the search results from different industries on the same preset ranking factor are different, for example, the catering industry is sensitive to the scene matching degree, and the user usually wants to eat nearby, while the photography industry is not sensitive to the scene matching degree, and the user is usually not in the distance of the photography store, so that, in practice, for each preset ranking factor in the preset ranking factors, the search intention and the search industry corresponding to the search content input by the user can be queried, and based on a linear function, the search intention, the search industry and each preset ranking factor are trained respectively to obtain a plurality of factor weights. Specifically, the training of the factor weight may be performed based on the following equation 2:

equation 2: w = f (intersection _ id, trade _ id, θ)

Wherein w is a factor weight; f is used to indicate the linear function employed; the interaction _ id is a search intention, and specifically can search attributes such as a brand, a type, content, an address and the like of the content; trade _ id is a search industry such as catering, medicine, retail, entertainment, fitness, etc.; and theta is a preset ordering factor of the current training. In order to continuously optimize the factor weight and make the factor weight conform to the constraint of the value interval of each information group, the following formula 3 may be adopted to calculate the calculated factor weight, so that the weight of the factor weight corresponding to the information group is the optimal weight.

Equation 3: w' = argmaxw (∑ τ R/| τ |)

W' is the optimal weight, argmaxw indicates Σ τ R/| τ | for maximizing w, R is the result of weighted summation of weights for each factor obtained from high to low positions, and τ indicates the round of weighted summation.

By repeatedly performing the above-described process of calculating the relevance scores of the search results, the relevance scores can be calculated for each search result, respectively, to obtain the relevance score of each search result.

After the relevance score of each search result is obtained through calculation, all search results included in the information groups are sorted according to the sequence of the relevance scores from large to small for each information group in the plurality of information groups, and the sorted information groups are obtained, so that the intra-group sorting of the information groups is realized. And repeatedly executing the sorting process to respectively calculate the relevance scores for the search results contained in each of the plurality of information groups and sort the relevance scores to obtain a plurality of sorted information groups. Specifically, referring to the content of setting the information group names for the information groups in step 202, it is known that there are good or bad groups for a plurality of information groups, and therefore, after the group sorting of the information groups is completed, the sorted information groups need to be combined according to the order of the group levels corresponding to the information groups from high to low to obtain an intermediate result queue. For example, assuming that the information group "good" is "ACDBE", the information group "good" is "FHIGJ", and the information group "poor" is "OLKMN", the resulting intermediate result queue is "ACDBE FHIGJ OLKMN".

204. And adjusting the sequence of the search results included in the intermediate result queue based on the relevance between the search results of the same order in the initial result queue and the intermediate result queue to obtain a target result queue.

In the embodiment of the present application, the search results have already formed an intermediate result queue through the processes in step 202 to step 203. However, there are intrinsic relations between the search results, and there is a bidirectional change in the user's intention during browsing, so in the embodiment of the present application, the intrinsic relations between the search results and the influence of the context environment of the intermediate result queue on the user are comprehensively considered, and the adjustment and rearrangement of the intermediate result queue are performed, so that the adjusted intermediate result queue is more reasonable, and a new target result queue is formed. The process of adjusting the intermediate result queue includes two steps, specifically as follows:

step one, comparing the initial result queue with the intermediate result queue, and adjusting the intermediate result queue.

When the initial result queue and the intermediate result queue are compared, for every two search results which are in the same order in the initial result queue and the intermediate result queue, the two search results are compared, the relevance between the two search results is determined, the sequence of the two search results is determined according to the relevance between the two search results, the intermediate result queue is adjusted, and the adjusted intermediate result queue is obtained. For example, assuming the initial result queue is "12345" and the intermediate result queue is "23541", a "1" in the initial result queue is compared to a "2" in the intermediate result queue and "21" is recorded when it is determined that "2" is better than "1". And then, continuously comparing the '2' in the initial result queue with the '3' in the intermediate result queue, and adding the '3' to the recorded content to obtain '321' when the '3' is determined to be superior to the '2', and repeating the steps until the comparison between the whole initial result queue and the intermediate result queue is finished, so that the adjusted intermediate result queue is the '32145'.

In the process of practical application, the internal relation between two search results can be identified through GRU (Gate recovery Unit) and Attention structure modeling, then the two search results are compared through a Pointer network, and the iteration is carried out to obtain an adjusted intermediate result queue.

And step two, evaluating the adjusted intermediate result queue and outputting a queue score.

In a search scenario, whether a user interacts with the last output result queue is greatly influenced by the context, in addition to the user and the search results themselves. Here, these two effects are mainly considered: one is the two-way change in the user's intent during browsing. For example, when the user browses the search results in a sliding manner, and when the user browses the search result of the 7 th digit, the user wants to remember what the search result of the 3 rd digit is, the user will look up the 3 rd digit again, that is, the user does not browse backwards in a tasteless manner, and may also look back. Generally, in browsing a certain page, in addition to the change of intentions occurring when browsing search results sequentially, the subsequent search results also have an influence on the intentions of the user, especially in the case of a two-tier information flow. The other is a collaborative relationship between stores, and this location-independent influence helps extract longer-term relationship dependencies. For example, suppose the user only likes a hot pot, so no matter how the hot pot and the rice noodles are sorted, the selection of the hot pot by the user is not influenced, and even if the rice noodles are arranged in front, the user can slide backwards to find the hot pot, so that the influence between two search results in the scene is avoided. And if the user likes the chafing dish and the rice noodles at the same time, the chafing dish and the rice noodles have strong internal connection, and the user can easily select which is arranged in the front. Therefore, only by capturing the influence of the context environment in the result list, the context perception can be really achieved.

Therefore, the search platform is provided with an evaluator, the evaluator is trained on the basis of the plurality of sample queues and indicates the grade of each sample queue in the plurality of sample queues, the evaluator can grade the adjusted middle result queue according to the incidence relation between the user intention and the search result, and then the result queue with the highest grade is selected as the final result queue in the follow-up process. Specifically, the evaluator may model a synergistic relationship between the user intention and the search result through Bi-LSTM (Bidirectional-Long Short Term Memory) and Self-attention mechanism, and the evaluator indicates that there are a large number of scores corresponding to the sample queues, so that the adjusted intermediate result queues are input to the evaluator for evaluation to obtain queue scores output by the evaluator, and the adjusted intermediate result queues are labeled by using the queue scores. For example, assuming that the adjusted intermediate result queue is "34521", and the score corresponding to the sample queue "34521" in the evaluator is 10, the queue score of the adjusted intermediate result queue is 10.

And repeating the comparison process in the first step and the grading process in the second step, comparing the adjusted intermediate result queue with the initial result queue, and readjusting the adjusted intermediate result queue until the adjustment times reach the preset cycle times, so that queue grading with the number meeting the preset cycle times can be obtained. And then, extracting a target queue score with the highest queue score and a target intermediate result queue labeled by the target queue score from the queue scores with the number meeting the preset turn times. Considering that the length of the output result list needs to be limited, a second preset number may need to be determined, search results of which the number meets the second preset number are intercepted at the head of the target intermediate result queue, and the search results of which the number meets the second preset number are used as the target result queue.

It should be noted that, in the embodiment of the present application, a second preset number is applied when determining the target result queue, and in the actual application process, when the adjusted intermediate result queue is obtained in step one, the adjusted intermediate result queue may be intercepted according to the second preset number, so as to reduce the pressure of subsequently creating an evaluator and performing queue evaluation based on the evaluator. Moreover, the second preset number can be set according to the length of the information group 'excellent', and can be consistent with the length of the information group 'excellent' or be larger than the length of the information group 'excellent', so that search results in the information group 'excellent' can be considered, meanwhile, the search results can be properly extended to the information group 'excellent', and the excellent search results are prevented from being filtered.

205. And adjusting the search results included in the target result queue according to the regulation and control requirements.

In the embodiment of the application, some regulation and control requirements are set along with the release of activities, the change of merchant appeal and the like in some search scenes, such as scattering of search results of the same brand, distance order preservation enhancement, weighting of newly-added search results, anti-cheating and the like. Therefore, in the practical application process, the preset regulation and control strategy needs to be queried, the regulation and control requirements of the preset regulation and control strategy are determined, the search results included in the target result queue are sequentially adjusted according to the regulation and control requirements, the regulated and controlled target result queue is obtained, and the regulated and controlled target result queue is output subsequently.

The preset regulation and control strategy can be divided into three types, the first type is a service strategy, such as search result scattering, distance order preserving enhancement and the like of the same brand. And the second type is an order-preserving strategy, and the search content in the target result queue is regulated and controlled according to the real-time flow in the search platform. And the third is an anti-cheating strategy, which is accessed to an external anti-cheating system and intervenes in the target result queue based on the anti-cheating system.

206. And outputting the target result queue.

In the embodiment of the application, after the target result queue is obtained, the target result queue is output. Thus, through the process, the multidimensional grouping, the in-group sorting, the queue rearrangement and the service order regulation of the initial result queue are realized, the sorting is carried out according to the excellent degree of the search results, and the excellent search results are arranged in the front. In addition, in the method, the in-group sequencing and the queue rearrangement are realized by learning based on a large-scale deep machine learning model, and the complex and high-dimensional parameter space can be faced, so that the method provided by the application has strong popularization in various scenes.

Further, as a specific implementation of the method shown in fig. 1, an embodiment of the present application provides a search result output device, and as shown in fig. 3A, the device includes: an obtaining module 301, a dividing module 302, a sorting module 303, an adjusting module 304 and an outputting module 305.

The obtaining module 301 is configured to obtain an initial result queue, where the initial result queue includes a plurality of search results related to search content determined based on the search content input by a user;

the dividing module 302 is configured to divide the plurality of search results into a plurality of information groups according to a matching degree between each search result of the plurality of search results and the search content and the geographic location where the user is located;

the sorting module 303 is configured to sort the search results included in each of the plurality of information groups according to a plurality of preset sorting factors to obtain an intermediate result queue;

the adjusting module 304 is configured to adjust an order of search results included in the intermediate result queue based on a correlation between the initial result queue and a search result of the same rank in the intermediate result queue, so as to obtain a target result queue;

the output module 305 is configured to output the target result queue.

In a specific application scenario, the obtaining module 301 is configured to receive the search content input by the user; analyzing the search content, and inquiring the plurality of search results related to the search content; performing Click Through Rate (CTR) estimation on the plurality of search results, and outputting a plurality of estimated click rates of the plurality of search results; and sequencing the plurality of search results according to the plurality of estimated click rates to obtain the initial result queue.

In a specific application scenario, the dividing module 302 is configured to determine, for each search result in the plurality of search results, a semantic matching degree and a scene matching degree of the each search result, where the semantic matching degree indicates a semantic correlation degree between the search result and the search content, and the scene matching degree indicates a correlation degree between the search result and the search scene; determining a plurality of semantic matching degree value intervals and a plurality of scene matching degree value intervals; comparing the semantic matching degree and the scene matching degree of each search result with the semantic matching degree value interval and the scene matching degree value intervals, and dividing the search results with the semantic matching degree and the scene matching degree in the same value interval into the same group to obtain a plurality of information groups.

In a specific application scenario, the dividing module 302 is configured to, for each search result in the plurality of search results, query the semantic matching degree between the search result and the search content; determining the geographic position of the user when the user inputs the search content and the store positions of stores related to the search result; and calculating the position distance between the geographic position and the store position, and acquiring the scene matching degree indicated by the position distance.

In a specific application scenario, the partitioning module 302 is configured to query a preset partitioning standard, and extract the multiple semantic matching degree value intervals and the multiple scene matching degree value intervals from the preset partitioning standard; or counting a first preset number of sample parameters, performing semantic training on the first preset number of sample parameters based on a machine learning depth model to obtain a plurality of semantic matching degree values, constructing a plurality of semantic matching degree value intervals according to the size relationship among the plurality of semantic matching degree values, performing scene training on the first preset number of sample parameters based on the machine learning depth model to obtain a plurality of scene matching degree values, and constructing the plurality of scene matching degree value intervals according to the size relationship among the plurality of scene matching degree values.

In a specific application scenario, the ranking module 303 is configured to calculate, for each information group in the plurality of information groups, a relevance score of each search result included in each information group based on the plurality of preset ranking factors; sequencing all search results included in the information group according to the sequence of the relevance scores from large to small to obtain the sequenced information group; calculating and sequencing relevance scores for search results included in each of the plurality of information groups respectively to obtain a plurality of sequenced information groups; and combining the sequenced information groups according to the sequence of the grouping grades corresponding to the information groups from high to low to obtain the intermediate result queue.

In a specific application scenario, the ranking module 303 is configured to, for each search result included in the information group, query a plurality of factor scores of the search result on the plurality of preset ranking factors; acquiring a plurality of factor weights corresponding to the preset ranking factors, and performing weight calculation on the factor scores based on the factor weights to obtain the relevance scores of the search results; and respectively calculating the relevance score for each search result to obtain the relevance score of each search result.

In a specific application scenario, the ranking module 303 is configured to query, for each preset ranking factor in the plurality of preset ranking factors, a search intention and a search industry corresponding to search content input by a user; and training the search intention, the search industry and each preset ranking factor respectively based on a linear function to obtain the multiple factor weights.

In a specific application scenario, the adjusting module 304 is configured to compare every two search results in the initial result queue and the intermediate result queue at the same rank to determine a relevance between the two search results; determining the sequence of the two search results and adjusting the intermediate result queue according to the relevance between the two search results to obtain the adjusted intermediate result queue; inputting the adjusted intermediate result queue to an evaluator for evaluation to obtain a queue score output by the evaluator, and labeling the adjusted intermediate result queue by using the queue score, wherein the evaluator is trained based on a plurality of sample queues and indicates the score of each sample queue in the plurality of sample queues; repeatedly executing the comparison process, comparing the adjusted intermediate result queue with the initial result queue and readjusting the adjusted intermediate result queue until the adjustment times reach the preset rotation times, and obtaining queue scores of which the number meets the preset rotation times; extracting a target queue score with the highest queue score and a target intermediate result queue labeled by the target queue score from the queue scores with the number meeting the preset cycle times; and determining a second preset number, intercepting the search results of which the number meets the second preset number at the head of the target intermediate result queue, and taking the search results of which the number meets the second preset number as the target result queue.

In a specific application scenario, as shown in fig. 3B, the apparatus further includes: and a query module 306.

The query module 306 is configured to query a preset regulation and control strategy and determine a regulation and control requirement of the preset regulation and control strategy;

the adjusting module 304 is further configured to sequentially adjust the search results included in the target result queue according to the adjustment and control requirement, so as to obtain the adjusted and controlled target result queue;

the output module 305 is further configured to output the regulated target result queue.

According to the device provided by the embodiment of the application, after the initial result queue is obtained according to the search content of the user, the search results in the initial result queue are divided into a plurality of information groups according to the matching degree between each search result in the initial result queue and the search content and the geographic position of the user, the search results included in the information groups are sorted in groups according to a plurality of preset sorting factors, and the sorted information groups are integrated to obtain the intermediate result queue. And then, based on the relevance between the initial result queue and the search result of the same rank in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain and output a target result queue, so that the scene where the user searches and the relevance between the search results are considered in the sorting process of the target result queue, the diversified sorting of the search results is realized, the search results arranged in the front in the target result queue are ensured to be attached to the search requirements of the user, the accuracy of the output search results is improved, and the universality is better.

It should be noted that other corresponding descriptions of the functional units related to the search result output device provided in the embodiment of the present application may refer to the corresponding descriptions in fig. 1 and fig. 2A, and are not repeated herein.

In an exemplary embodiment, referring to fig. 4, there is further provided a device including a bus, a processor, a memory, and a communication interface, and further including an input-output interface and a display device, wherein the respective functional units may perform communication with each other through the bus. The memory stores computer programs, and the processor is used for executing the programs stored in the memory and executing the search result output method in the embodiment.

A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the search result output method.

Through the above description of the embodiments, those skilled in the art will clearly understand that the present application can be implemented by hardware, and also by software plus a necessary general hardware platform. Based on such understanding, the technical solution of the present application may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the implementation scenarios of the present application.

Those skilled in the art will appreciate that the figures are merely schematic representations of one preferred implementation scenario and that the blocks or flow diagrams in the figures are not necessarily required to practice the present application.

Those skilled in the art will appreciate that the modules in the devices in the implementation scenario may be distributed in the devices in the implementation scenario according to the description of the implementation scenario, or may be located in one or more devices different from the present implementation scenario with corresponding changes. The modules of the implementation scenario may be combined into one module, or may be further split into a plurality of sub-modules.

The above application serial numbers are for description purposes only and do not represent the superiority or inferiority of the implementation scenarios.

The above disclosure is only a few specific implementation scenarios of the present application, but the present application is not limited thereto, and any variations that can be made by those skilled in the art are intended to fall within the scope of the present application.

Claims

1. A search result output method, comprising:

sorting the search results included in each of the information groups according to a plurality of preset sorting factors to obtain an intermediate result queue, wherein the search results in the information groups are sorted in a descending order of the relevance scores calculated based on the preset sorting factors, and the intermediate result queue is obtained by combining the sorted information groups in a descending order of the grouping grades;

based on the relevance between the initial result queue and the search result of the same level in the intermediate result queue, adjusting the sequence of the search results included in the intermediate result queue to obtain a target result queue, wherein the relevance between the two search results of the same level is identified through GRU and Attention structural modeling, and the sequence of the two search results is determined according to the relevance and is adjusted;

and outputting the target result queue.

2. The method of claim 1, wherein obtaining an initial result queue comprises:

receiving the search content input by the user;

3. The method of claim 1, wherein the dividing the plurality of search results into a plurality of information groups according to the matching degree between each search result of the plurality of search results and the search content and the geographic location of the user when searching comprises:

for each search result in the plurality of search results, determining a semantic matching degree and a scene matching degree of the each search result, wherein the semantic matching degree indicates the semantic relevance degree of the search result and the search content, and the scene matching degree indicates the relevance degree of the search result and the geographic position;

4. The method of claim 3, wherein the determining the semantic matching degree and the scene matching degree of each search result comprises:

5. The method of claim 3, wherein determining the plurality of semantic matching degree intervals and the plurality of scene matching degree intervals comprises:

6. The method of claim 1, wherein said sorting search results included in each of said plurality of information groups by a plurality of predetermined sorting factors to obtain an intermediate result queue comprises:

7. The method of claim 6, wherein said calculating a relevance score for each search result included in said each information group based on said plurality of preset ranking factors comprises:

8. The method according to claim 7, wherein the obtaining a plurality of factor weights corresponding to the plurality of preset ranking factors comprises:

9. The method of claim 1, wherein the adjusting the order of the search results included in the intermediate result queue based on the correlation between the initial result queue and the search result of the same rank in the intermediate result queue to obtain a target result queue comprises:

10. A search result output apparatus characterized by comprising:

the sorting module is used for sorting the search results included by each of the information groups according to a plurality of preset sorting factors to obtain an intermediate result queue, wherein the search results in the information groups are sorted according to a sequence from large to small of the correlation scores calculated based on the preset sorting factors, and the intermediate result queue is obtained by combining the sorted information groups according to a sequence from high to low of the grouping grades;

the adjusting module is used for adjusting the sequence of the search results included in the intermediate result queue based on the relevance between the initial result queue and the search result of the same order in the intermediate result queue to obtain a target result queue, wherein the relevance between the two search results of the same order is identified through GRU and Attention structure modeling, and the sequence of the two search results is determined according to the relevance and is adjusted;

and the output module is used for outputting the target result queue.