Search recommendation method and device, electronic equipment and computer storage medium

Abstract

The invention discloses a search recommendation method and device, electronic equipment and a computer storage medium. Accurate recommendation of addresses can be achieved. The method comprises the following steps: acquiring multi-dimensional search related information of a user; starting a first link according to the multi-dimensional search related information; the first link comprises a plurality of first task nodes, each first task node is used for executing different stage tasks, and concurrent streaming data transmission is carried out among the first task nodes; and processing the multi-dimensional search related information through the first link to obtain a first search recommendation result. By adopting the method, on the basis of ensuring that the multi-dimensional search related information is effectively processed, the total duration of search recommendation is effectively shortened, and the response speed of search recommendation is improved.

Description

Technical Field

[0001] This application relates to the technical field of search and recommendation, and particularly to a search and recommendation method, apparatus, electronic device, and computer storage medium. Background Art

[0002] The search and recommendation process is usually divided into multiple stages, such as recall, rough ranking, and fine ranking. In related technologies, a serial execution link is used to execute each stage in sequence, that is, the tasks of each stage need to wait until the tasks of the previous stage are completely finished before they can start. Therefore, if a long-tail problem occurs during the execution of any stage (that is, the execution time of some tasks is abnormally extended), it will cause the total time of search and recommendation to increase, thereby reducing the response speed of search and recommendation. There is a need to provide a search and recommendation solution with a faster response speed. Summary of the Invention

[0003] Embodiments of this application provide a search and recommendation method, apparatus, electronic device, and computer storage medium, which can effectively improve the response speed of search and recommendation. The above technical solutions are as follows:

[0004] In a first aspect, embodiments of this application provide a search and recommendation method, including:

[0005] Obtain multi-dimensional search-related information of a user;

[0006] Start a first link according to the multi-dimensional search-related information; the first link includes multiple first task nodes, and each first task node is used to execute different stage tasks, and data is transmitted between each first task node in a concurrent streaming manner;

[0007] Process the multi-dimensional search-related information through the first link to obtain a first search and recommendation result.

[0008] In a possible implementation, the multiple first task nodes include at least one of a first recall node, a first screening node, and a first ranking node, and a merging node; the first ranking node includes a first fine ranking node; the merging node is used to collect the data flowing in within a first preset time period; the starting moment of the first preset time period is determined by the starting moment of the first link.

[0009] In a possible implementation, the multiple first task nodes include a first screening node and a merging node; the merging node is used to collect the data flowing in within a first preset time period; the starting moment of the first preset time period is determined by the starting moment of the first link;

[0010] Before starting the first link according to the multi-dimensional search-related information, the method further includes:

[0011] Determine whether there is a pre-determined screening threshold for each dimension of search-related information;

[0012] If there are predetermined screening values for the multi-dimensional search related information, perform the step of starting the first link according to the multi-dimensional search related information.

[0013] In a possible implementation, the multiple first task nodes further include a first recall node and a first fine ranking node;

[0014] Process the multi-dimensional search related information through the first link to obtain a first search recommendation result, including:

[0015] Recall at least one first candidate object through the first recall node according to the multi-dimensional search related information, and after recalling any first candidate object, transmit any first candidate object to the first screening node;

[0016] Screen the incoming first candidate objects through the first screening node to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmit any first object to be recommended to the first fine ranking node;

[0017] Perform fine ranking scoring on the incoming first objects to be recommended through the first fine ranking node to obtain a fine ranking scoring result corresponding to at least one first object to be recommended, and after obtaining a fine ranking scoring result corresponding to any first object to be recommended, transmit any first object to be recommended and its corresponding fine ranking scoring result to the merging node;

[0018] Collect the incoming first objects to be recommended and their corresponding fine ranking scoring results within a first preset time period through the merging node;

[0019] Determine the first search recommendation result based on the first objects to be recommended and their corresponding fine ranking scoring results collected by the merging node.

[0020] In a possible implementation, the first screening node includes a first rough ranking scoring model and a dynamic threshold model;

[0021] Screen the incoming first candidate objects through the first screening node to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmit any first object to be recommended to the first fine ranking node, including:

[0022] Perform rough ranking scoring on the incoming first candidate objects through the first rough ranking scoring model to obtain a rough ranking scoring result corresponding to each first candidate object;

[0023] The dynamic threshold model filters each incoming first candidate object based on the rough ranking score results corresponding to each incoming first candidate object and the screening thresholds corresponding to the search-related information in each dimension, to obtain at least one first object to be recommended. After obtaining any first object to be recommended, any first object to be recommended is transmitted to the first fine ranking node.

[0024] In a possible implementation, the first screening node further includes a deduplication table, which is used to record the incoming first candidate objects.

[0025] Before the dynamic threshold model filters each incoming first candidate object based on the rough ranking score results corresponding to each first candidate object and the screening thresholds corresponding to the search-related information in each dimension, to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, any first object to be recommended is transmitted to the first fine ranking node, the method further includes:

[0026] Determine whether the currently incoming first candidate object has been recorded in the deduplication table.

[0027] If so, discard the currently incoming first candidate object.

[0028] If not, retain the currently incoming first candidate object.

[0029] In a possible implementation, the above method further includes:

[0030] Determine whether the merging node has collected a preset number of first objects to be recommended within a second preset time period.

[0031] If the merging node has not collected a preset number of first objects to be recommended within a second preset time period, start a backup link.

[0032] Process the search-related information in multiple dimensions through the backup link according to a preset fallback strategy to obtain a second search recommendation result.

[0033] In a possible implementation, determining the first search recommendation result based on each first object to be recommended collected by the merging node and its corresponding fine ranking score results includes:

[0034] Use at least one first object to be recommended with a higher fine ranking score result among the first objects to be recommended collected by the merging node as at least one search recommendation object to obtain a first search recommendation result.

[0035] In a possible implementation, before determining the first search recommendation result based on each first object to be recommended collected by the merging node and its corresponding fine ranking score results, the method further includes:

[0036] In the case where the first recommended object collected by the merging node does not include the specified recommended object, the specified recommended object is collected by the merging node; the specified recommended object is recalled by a pre-specified key recall path.

[0037] Based on each first recommended object collected by the merging node and its corresponding fine ranking score result, determine the first search recommendation result, including:

[0038] Use at least one first recommended object with a relatively high fine ranking score result among the first recommended objects collected by the merging node, and at least one specified recommended object with a relatively high fine ranking score result among the specified recommended objects collected by the merging node as at least one search recommended object to obtain the first search recommendation result.

[0039] In a possible implementation, after determining the first search recommendation result based on each first recommended object collected by the merging node and its corresponding fine ranking score result, the method further includes:

[0040] Re-rank the first search recommendation result to obtain the re-ranked first search recommendation result.

[0041] In a possible implementation, the above method further includes:

[0042] If there is no pre-determined screening threshold for each dimension of search-related information, start the second link according to the multi-dimensional search-related information; the second link includes multiple second task nodes, and each second task node is used to execute different stage tasks, and data is transmitted sequentially between each second task node.

[0043] Process the multi-dimensional search-related information through the second link to obtain the third search recommendation result.

[0044] In a possible implementation, the multiple second task nodes include at least one of a second recall node, a second sorting node, and a second screening node; the second sorting node includes a second fine ranking node.

[0045] In a possible implementation, the multiple second task nodes include a second recall node, a second screening node, and a second fine ranking node.

[0046] Processing the multi-dimensional search-related information through the second link to obtain the third search recommendation result includes:

[0047] The second recall node recalls at least one second candidate object according to the multi-dimensional search-related information, and after recalling at least one second candidate object, transmits at least one second candidate object to the second screening node.

[0048] Screen at least one second candidate object through a second screening node to obtain at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmit at least one second object to be recommended to a second fine-ranking node;

[0049] Perform fine-ranking scoring on at least one second object to be recommended through the second fine-ranking node to obtain the fine-ranking scoring results corresponding to each second object to be recommended;

[0050] Determine a third search recommendation result based on at least one second object to be recommended and its corresponding fine-ranking scoring results.

[0051] In a possible implementation, the second screening node includes a second rough-ranking scoring model;

[0052] Screening at least one second candidate object through the second screening node to obtain at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmitting at least one second object to be recommended to the second fine-ranking node includes:

[0053] Perform rough-ranking scoring on at least one second candidate object respectively through the second rough-ranking scoring model to obtain the rough-ranking scoring results corresponding to each second candidate object;

[0054] Use at least one second candidate object with a higher rough-ranking scoring result among at least one second candidate object as at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmit at least one second object to be recommended to the second fine-ranking node.

[0055] In a possible implementation, the screening threshold is determined by the rough-ranking scoring results corresponding to at least one second candidate object.

[0056] In a second aspect, an embodiment of the present application provides a search recommendation method, which is applied to a user terminal and includes:

[0057] Receive a search request from a user;

[0058] Send the search request to the server so that the server, in response to the search request, obtains the multi-dimensional search-related information of the user according to the search information carried in the search request, starts a first link according to the multi-dimensional search-related information, and processes the multi-dimensional search-related information through the first link to obtain a first search recommendation result;

[0059] Wherein, the first link includes a plurality of first task nodes, and each first task node is used to execute different stage tasks, and data is transmitted between each first task node in a concurrent streaming manner.

[0060] In a third aspect, an embodiment of the present application provides a search recommendation system, including: a user terminal and a server; wherein,

[0061] A merchant terminal, configured to receive a search request from a user and send the search request to a server;

[0062] A server, configured to, in response to the search request, obtain multi-dimensional search-related information of the user according to the search information carried in the search request, initiate a first link according to the multi-dimensional search-related information, process the multi-dimensional search-related information through the first link to obtain a first search recommendation result, and send the first search recommendation result to a user terminal; wherein, the first link includes a plurality of first task nodes, and each first task node is configured to execute different stage tasks, and data is transmitted between the first task nodes in a concurrent streaming manner;

[0063] The merchant terminal is further configured to receive the first search recommendation result and perform display based on the first search recommendation result.

[0064] In a fourth aspect, an embodiment of the present application provides a search recommendation device, including:

[0065] An acquisition module, configured to acquire multi-dimensional search-related information of a user;

[0066] A start module, configured to initiate a first link according to the multi-dimensional search-related information; the first link includes a plurality of first task nodes, and each first task node is configured to execute different stage tasks, and data is transmitted between the first task nodes in a concurrent streaming manner;

[0067] A processing module, configured to process the multi-dimensional search-related information through the first link to obtain a first search recommendation result.

[0068] In a fifth aspect, an embodiment of the present application provides a search recommendation device, which is applied to a user terminal and includes:

[0069] A first receiving module, configured to receive a search request from a user;

[0070] A sending module, configured to send the search request to a server, so that the server, in response to the search request, obtains multi-dimensional search-related information of the user according to the search information carried in the search request, initiates a first link according to the multi-dimensional search-related information, processes the multi-dimensional search-related information through the first link to obtain a first search recommendation result, and sends the first search recommendation result to the user terminal; wherein, the first link includes a plurality of first task nodes, and each first task node is configured to execute different stage tasks, and data is transmitted between the first task nodes in a concurrent streaming manner;

[0071] A second receiving module, configured to receive the first search recommendation result;

[0072] A display module, configured to perform display based on the first search recommendation result.

[0073] Sixth aspect, an embodiment of the present application provides an electronic device, including: a processor and a memory; wherein, the memory stores a computer program, and when the processor executes the computer program, the method steps provided in the first aspect or the second aspect of the embodiments of the present application are implemented.

[0074] Seventh aspect, an embodiment of the present application provides a computer storage medium, which stores multiple instructions, and the instructions are adapted to be loaded and executed by a processor to implement the method steps provided in the first aspect or the second aspect of the embodiments of the present application.

[0075] The above search recommendation method, device, electronic device and computer storage medium can ensure that the multi-dimensional search-related information is effectively processed through multiple different stages by obtaining the multi-dimensional search-related information of the user and starting a first link including multiple first task nodes according to the multi-dimensional search-related information, and each first task node is used to execute different stage tasks; in addition, based on the mechanism of concurrent streaming data transmission between the first task nodes, the multi-dimensional search-related information is processed through the first link to obtain a first search recommendation result, which can effectively shorten the total duration of the search recommendation and improve the response speed of the search recommendation on the basis of ensuring the effective processing of the multi-dimensional search-related information. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0077] Figure 1 A data flow diagram of a search recommendation link provided for an exemplary embodiment of the present application;

[0078] Figure 2 An architecture diagram of a search recommendation system provided for an exemplary embodiment of the present application;

[0079] Figure 3 A flowchart of a search recommendation method provided for an exemplary embodiment of the present application;

[0080] Figure 4 A flowchart of another search recommendation method provided for an exemplary embodiment of the present application;

[0081] Figure 5 A flowchart of another search recommendation method provided for an exemplary embodiment of the present application;

[0082] Figure 6Schematic flowchart of another search and recommendation method provided by an exemplary embodiment of the present application;

[0083] Figure 7 Schematic structural diagram of a search and recommendation device provided by an exemplary embodiment of the present application;

[0084] Figure 8 Schematic structural diagram of another search and recommendation device provided by an exemplary embodiment of the present application;

[0085] Figure 9 Schematic structural diagram of an electronic device provided by an exemplary embodiment of the present application. Detailed implementation manners

[0086] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

[0087] In the description of the present application, it should be understood that terms such as "first" and "second" are only used for descriptive purposes, and cannot be construed as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances. In addition, in the description of the present application, unless otherwise specified, "a plurality of" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. These three situations. The character " / " generally represents an "or" relationship between the associated objects before and after.

[0088] In order to describe the technical solutions of the embodiments of the present application more clearly, some concepts in the present application are described in detail before the description for better understanding of the solution.

[0089] Concurrency: In a multi-task environment, concurrency enables multiple tasks to be performed simultaneously. In the embodiments of the present application, concurrency specifically means that multiple task nodes can independently process different data items (such as doc documents) in the same time period, improving the efficiency and response speed of search and recommendation.

[0090] Streaming transmission: It means that data items can be processed one by one, instead of waiting for all data to be ready at once before starting to process. This method allows for faster response to partial results, thus better adapting to search and recommendation scenarios with high real-time requirements.

[0091] Please refer to Figure 1, which is a schematic data flow diagram of a search and recommendation link provided by an exemplary embodiment of the present application. The search and recommendation link includes a query processing node 10, a first recall node 11, a first screening node 12, a first fine ranking node 13, a merging node 14, and a re-ranking node 15. Among them, the first recall node 11, the first screening node 12, the first fine ranking node 13, and the merging node 14 constitute a concurrent streaming execution link (i.e., the first link in the embodiment of the present application). Data is transmitted concurrently and streamingly between the first recall node 11, the first screening node 12, the first fine ranking node 13, and the merging node 14.

[0092] Specifically, the query processing node 10 is used to receive the user's search request and process the search information carried by the search request to generate multi-dimensional search-related information. The first recall node 11 includes a recall engine (not shown in the figure), which is used to start N (N is a positive integer) recall paths to recall at least one first candidate object according to the multi-dimensional search-related information generated by the query processing node 10 and a preset recall strategy, so as to execute the tasks in the recall stage. The first screening node 12 includes a deduplication table, a dynamic threshold model, a rough ranking scoring model, and a feature statistics model, which are used to perform dynamic threshold screening on the at least one first candidate object recalled to obtain at least one first object to be recommended. The first fine ranking node 13 includes at least one fine ranking scoring model, such as Figure 1 the product fine ranking scoring model, the store fine ranking scoring model, etc. shown, which are used to perform fine ranking scoring on the at least one first object to be recommended screened out to obtain the fine ranking scoring result corresponding to the at least one first object to be recommended. The merging node 14 is used to collect the first objects to be recommended that have completed fine ranking scoring within a preset time period to obtain a first search and recommendation result. The re-ranking node 15 is used to re-rank the first search and recommendation result to obtain the re-ranked first search and recommendation result.

[0093] Exemplarily, the working process of the above search and recommendation link is as follows: The query processing node 10 performs query processing on the user's original search information and the user's context information to obtain the user's multi-dimensional search-related information, such as search keyword information, geographical location information at the time of search, timestamp information at the time of search, etc., and sends the multi-dimensional search-related information to the first recall node 11 to start the concurrent streaming execution link.

[0094] After the concurrent streaming execution link, the first recall node 11 starts N recall paths to recall at least one first candidate object according to the multi-dimensional search-related information and the preset recall strategy, and after recalling any first candidate object, transmits any first candidate object to the first screening node 12.

[0095] The first screening node 12 screens each of the incoming first candidate objects to obtain at least one first object to be recommended. After obtaining any first object to be recommended, the first object to be recommended is transmitted to the first fine-ranking node 13.

[0096] The first fine-ranking node 13 performs fine-ranking scoring on each of the incoming first objects to be recommended to obtain fine-ranking scoring results corresponding to at least one first object to be recommended. After obtaining the fine-ranking scoring result corresponding to any first object to be recommended, the first object to be recommended and its corresponding fine-ranking scoring result are transmitted to the merging node 14.

[0097] The merging node 14 collects each of the incoming first objects to be recommended and their corresponding fine-ranking scoring results within the first preset time period, and transmits all the collected first objects to be recommended and their corresponding fine-ranking scoring results to the re-ranking node 15 at the end of the first preset time period.

[0098] The re-ranking node 15 re-ranks the first search recommendation result according to a preset diversity strategy to obtain a re-ranked first search recommendation result. The search recommendation link uses the first search recommendation result re-ranked by the re-ranking node 15 as the final output result.

[0099] Please refer to Figure 2 , which is a schematic diagram of the architecture of a search recommendation system provided by an exemplary embodiment of the present application. Among them, the user terminal 20 logged in to the online search platform communicates with the server 21 corresponding to the online search platform through the network. The data storage system can store the data that the server 21 needs to process. The data storage system can be integrated on the server 21, or can be placed on the cloud or other network servers.

[0100] In some possible embodiments, the search recommendation method provided by the present application can be jointly executed by the user terminal 20 and the server 21. Correspondingly, the search recommendation device can also be respectively set in the merchant terminal 30 and the server 31. Optionally, the user terminal 20 receives a search request input by the user and sends the search request to the server 21. The server 21 receives the search request sent by the user terminal 20, starts the first link according to the search information carried in the search request, processes the multi-dimensional search-related information through the first link to obtain a first search recommendation result, and sends the first search recommendation result to the user terminal 20. Among them, the first link includes a plurality of first task nodes, and each first task node is used to execute different stage tasks, and data is transmitted between each first task node in a concurrent streaming manner. The user terminal 20 receives the first search recommendation result sent by the server 21 and displays it based on the first search recommendation result.

[0101] In some possible embodiments, the search recommendation method provided by this application may be executed by the user terminal 20. Correspondingly, the search recommendation device may also be disposed in the user terminal 20.

[0102] In some possible embodiments, the search recommendation method provided by this application may be executed by the server 21. Correspondingly, the search recommendation device may also be disposed in the server 21.

[0103] It should be noted that the above online search platform may be, but is not limited to, an online shopping platform, an online food ordering platform, an online navigation platform, etc. The user terminal 20 may be, but is not limited to, various smart phones, tablet computers, personal computers, laptop computers, Internet of Things devices, and portable wearable devices, etc. The Internet of Things devices may be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 21 may be implemented by an independent server or a server cluster composed of multiple servers.

[0104] In one embodiment, as Figure 3 shown, a search recommendation method is provided. Taking the method applied to the Figure 1 server 21 as an example for description, the method includes the following steps:

[0105] S301: Obtain multi-dimensional search-related information of the user.

[0106] Among them, the multi-dimensional search-related information may include, but is not limited to, search keyword information, geographical location information at the time of search, timestamp information at the time of search, etc.

[0107] Optionally, the user terminal receives a search request input by the user and sends the search request to the server 21; the search request carries search information, and the search information may include, but is not limited to, the user's original search information and the user's context information. The server 21 receives the search request sent by the user terminal and obtains the multi-dimensional search-related information of the user according to the search information carried in the search request.

[0108] Specifically, the user's original search information may include, but is not limited to, at least one of the following: the user's search text information, the user's search picture information, the user's search audio information. The user's context information may include, but is not limited to: geographical location information at the time of user search, timestamp information at the time of user search, the user's account identification information, etc. After receiving the search request sent by the user terminal, the server 21 performs query processing on the user's original search information and the user's context information through a search engine to obtain the multi-dimensional search-related information of the user, such as search keyword information, geographical location information at the time of search, timestamp information at the time of search, etc.

[0109] In this embodiment, the server queries and processes the user's original search information and the user's context information through a search engine to obtain the user's multi-dimensional search-related information, which can ensure the richness and rationality of the multi-dimensional search-related information, provide a reliable information source for the subsequent processing stage, and help improve the effect of search recommendations.

[0110] S302: Start the first link according to the multi-dimensional search-related information; the first link includes multiple first task nodes, and each first task node is used to execute different stage tasks, and data is transmitted in a concurrent streaming manner between the first task nodes.

[0111] Among them, the first link is a concurrent streaming execution link.

[0112] Optionally, after the search engine of the server 21 queries and processes the user's original search information and the user's context information to obtain the user's multi-dimensional search-related information, the multi-dimensional search-related information is sent to the entry node (such as the first recall node) of the first link, thereby starting the first link.

[0113] Exemplarily, the multiple first task nodes of the first link include a merge node (join node), and at least one of the following: a first recall node for executing the recall stage task, a first screening node for executing the screening stage task, and a first sorting node for executing the sorting stage task. It can be understood that when the user searches for a certain geographical location information through an online navigation platform, since there are fewer locations corresponding to the geographical location information, the first link may only include a first recall node, a first fine sorting node, and a merge node; when the user searches for a certain commodity through an online shopping platform, since there are many commodity types and merchants, the first link may include a first recall node, a first screening node, a first fine sorting node, and a merge node.

[0114] It should be noted that the above first sorting node may include, but is not limited to, a first rough sorting node for executing the rough sorting stage task, a first fine sorting node for executing the fine sorting stage task, etc. The above merge node may be the last node in the first link, and it is used to collect the data flowing in within the first preset time period. The server 21 can determine the first search recommendation result according to the data collected by the merge node within the first preset time period, thereby effectively controlling the total time consumption of the entire link. The start time of the above first preset time period may be determined by, but is not limited to, the start time of the first link.

[0115] In this embodiment, by using the concurrent streaming execution link including the merge node as the first link, the data items collected within the first preset time period can be preferentially returned to the user without waiting for all the data items to be returned, thereby effectively controlling the total time consumption of the entire link, shortening the total duration of search recommendations, and improving the response speed of search recommendations.

[0116] S303: Process the multi-dimensional search related information through the first link to obtain the first search recommendation result.

[0117] Optionally, the server 21 processes the multi-dimensional search related information such as the search keyword information, the geographical location information at the time of search, and the timestamp information at the time of search through the first link to obtain the first search recommendation result. After obtaining the first search recommendation result, the server 21 can directly send the first search recommendation result to the user terminal so that the user terminal receives the first search recommendation result and displays the first search recommendation result on the user interface; or can continue to further process the first search recommendation result (such as re-ranking, etc.) and then send it to the user terminal for display.

[0118] Exemplarily, when the first link includes a first recall node, a first ranking node, and a merging node, the server 21 processes the multi-dimensional search related information through the first link to obtain the first search recommendation result, which may specifically include: recalling at least one first candidate object by the first recall node according to the multi-dimensional search related information, and after recalling any first candidate object, transmitting any first candidate object to the first fine-ranking node; performing fine-ranking scoring on each incoming first object to be recommended by the first fine-ranking node to obtain the fine-ranking scoring results corresponding to each first object to be recommended, and after obtaining the fine-ranking scoring result corresponding to any first object to be recommended, transmitting any first object to be recommended and its corresponding fine-ranking scoring result to the merging node; collecting each first object to be recommended and its corresponding fine-ranking scoring results flowing in within the first preset time period by the merging node; determining the first search recommendation result based on each first object to be recommended and its corresponding fine-ranking scoring results collected by the merging node.

[0119] It should be noted that this embodiment only describes a preferred embodiment of the present application. The first link may further include other multiple first task nodes, such as including a first recall node, a first screening node, a first fine-ranking node, and a merging node, etc. The embodiments of the present application do not limit the manner in which the above-mentioned first task nodes form the first link. In addition, for the case where the first link includes a first recall node, a first screening node, a first ranking node, and a merging node, the present application will be described in detail in subsequent embodiments and will not be elaborated here.

[0120] In this embodiment, the server processes the multi-dimensional search related information through the first link to obtain the first search recommendation result, which can realize the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism, thereby improving the execution efficiency of the entire link, effectively shortening the total duration of search recommendation, and improving the response speed of search recommendation.

[0121] The above search recommendation method obtains multi-dimensional search-related information of a user, and starts a first link including a plurality of first task nodes according to the multi-dimensional search-related information. Each first task node is used to execute different stage tasks, which can ensure that the multi-dimensional search-related information is effectively processed through multiple different stages. In addition, based on the mechanism of concurrent streaming data transmission between the first task nodes, the multi-dimensional search-related information is processed through the first link to obtain a first search recommendation result, which can effectively shorten the total duration of search recommendation and improve the response speed of search recommendation on the basis of ensuring the effective processing of the multi-dimensional search-related information.

[0122] In one embodiment, as Figure 4 shown, another search recommendation method is provided. Taking the method applied to the Figure 1 server 21 as an example, the method includes the following steps:

[0123] S401: Obtain multi-dimensional search-related information of a user.

[0124] Specifically, S401 is the same as S301, which will not be elaborated here.

[0125] S402: Determine whether there is a predetermined screening threshold for each dimension of search-related information. If yes, execute S403; if not, execute S405.

[0126] Among them, the screening threshold corresponding to each dimension of search-related information can be predetermined based on the rough ranking score result and screening result of the data items in the second link. For example, the screening threshold can be set to the lowest rough ranking score value corresponding to N data items (N is a preset positive integer) with relatively high rough ranking score results screened out by the second link. The second link is a serial execution link, which includes a plurality of second task nodes. Each second task node is used to execute different stage tasks, and the data is transmitted sequentially between the second task nodes.

[0127] It can be understood that when the server 21 performs search recommendation through the second link, each processing stage (such as the recall stage, the screening stage, and the fine ranking stage) is executed sequentially. That is, the current stage will only start the task logic after receiving all the data items output by the previous stage. Therefore, in the screening stage, the second link can screen out N data items with relatively high rough ranking score results by rough ranking and scoring each incoming data item and then integrating the rough ranking score results of all data items.

[0128] When the server 21 performs search and recommendation through the first link, data is streamed concurrently between each processing stage, that is, the task logic will be started as soon as any data item transmitted in the previous stage is received in the current stage. Therefore, the first link needs to use a limit value in the screening stage to determine whether each incoming data item can enter the next processing stage.

[0129] It is worth noting that, depending on the geographical location of the user when searching, when the server 21 performs search recommendations through the second link, the minimum values ​​of the rough ranking scores corresponding to the N data items with higher rough ranking scores that are screened out will often be different. For example, when a user searches for goods through an online ordering platform, if the geographical location of the user when searching is a core business district, where there is a large supply of goods nearby, then the minimum value of the rough ranking score will be relatively high; if the geographical location of the user when searching is a suburb, where there is a small supply of goods nearby, then the minimum value of the rough ranking score will be relatively low. Therefore, the boundary value required by the first link in the screening stage is not fixed, and needs to be dynamically determined in combination with the screening thresholds corresponding to the search-related information of each dimension each time the user initiates a search request. If there is no corresponding screening threshold for the search-related information of any dimension, the second link needs to be started for search recommendations.

[0130] S403: Start the first link according to the multi-dimensional search related information.

[0131] Optionally, after obtaining the multi-dimensional search related information of the user through query processing, the server 21 sends the multi-dimensional search related information to the entry node of the first link, thereby starting the first link.

[0132] The first link includes a plurality of first task nodes, each of which is used to perform a different stage task, and each of the first task nodes concurrently streams data. Optionally, the plurality of first task nodes include a first screening node and a merging node. The first screening node is used to perform the screening stage task according to a screening threshold predetermined according to the relevant information of each dimensional search. The merging node is used to collect the data flowing in within a first preset time period; the starting time of the first preset time period is determined by the start time of the first link.

[0133] It is worth noting that the above-mentioned merging node can be the exit node of the first link (i.e., the last task node of the first link), which is used to collect the data flowing in within the first preset time period. The server 21 can determine the first search recommendation result based on the data collected by the merging node within the first preset time period, thereby effectively controlling the total time consumption of the whole link. The starting time of the above-mentioned first preset time period can be determined by, but is not limited to, the start time of the first link.

[0134] In this embodiment, by determining whether there is a pre-determined screening threshold for each dimension of search-related information, when there is a screening threshold for each dimension of search-related information, the first link including the first screening node and the merging node is started. On the one hand, it is possible to perform dynamic threshold filtering in combination with the screening thresholds corresponding to each dimension of search-related information; on the other hand, it is possible to preferentially return the data items collected within the first preset time period to the user without waiting for all the data items to be returned, which not only ensures the quality of search recommendations, but also effectively controls the total time consumption of the entire link, shortens the total duration of search recommendations, and improves the response speed of search recommendations.

[0135] S404: Process the multi-dimensional search-related information through the first link to obtain the first search recommendation result.

[0136] Optionally, the server 21 processes multi-dimensional search-related information such as search keyword information, geographical location information at the time of search, and timestamp information at the time of search through the first link to obtain the first search recommendation result. After obtaining the first search recommendation result, the server 21 can directly send the first search recommendation result to the user terminal; or it can continue to further process the first search recommendation result (such as re-ranking, etc.) and then send it to the user terminal.

[0137] Exemplarily, when the first link includes a first recall node, a first screening node, and a merging node, the server 21 processes the multi-dimensional search-related information through the first link to obtain the first search recommendation result, which can specifically include: recalling at least one first candidate object according to the multi-dimensional search-related information through the first recall node, and after recalling any first candidate object, transmitting any first candidate object to the first screening node; screening each first candidate object flowing in through the first screening node to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmitting any first object to be recommended to the merging node; collecting each first object to be recommended flowing in within the first preset time period through the merging node; determining the first search recommendation result based on each first object to be recommended collected by the merging node.

[0138] In this embodiment, when there is a pre-determined screening threshold for each dimension of search-related information, the server processes the multi-dimensional search-related information through the first link to obtain the first search recommendation result, which can achieve concurrent processing of tasks in different stages and continuous transmission of data based on the concurrent streaming execution mechanism, thereby improving the execution efficiency of the entire link, effectively shortening the total duration of search recommendations, and improving the response speed of search recommendations.

[0139] S405: Start the second link according to the multi-dimensional search-related information.

[0140] Among them, the second link includes multiple second task nodes, each second task node is used to execute different stage tasks, and data is transmitted sequentially among the second task nodes. Optionally, the multiple second task nodes include at least one of a second recall node and a second sorting node, and a second screening node; the second sorting node includes a second fine sorting node.

[0141] Exemplarily, after the server 21 obtains the multi-dimensional search related information of the user through query processing, it sends the multi-dimensional search related information to the entry node of the second link (such as the second recall node), thereby starting the second link.

[0142] S406: Process the multi-dimensional search related information through the second link to obtain a third search recommendation result.

[0143] Optionally, the server 21 processes the multi-dimensional search related information such as search keyword information, geographical location information during search, and timestamp information during search through the second link to obtain a third search recommendation result. After the server 21 obtains the third search recommendation result, it can directly send the third search recommendation result to the user terminal; or it can continue to further process the third search recommendation result (such as re-sorting, etc.) and then send it to the user terminal.

[0144] Exemplarily, when the second link includes a second recall node and a second screening node, the server 21 processes the multi-dimensional search related information through the second link to obtain a second search recommendation result, which may specifically include: recalling at least one second candidate object according to the multi-dimensional search related information through the second recall node, and after recalling at least one second candidate object, transmitting at least one second candidate object to the second screening node; screening at least one second candidate object through the second screening node to obtain at least one second object to be recommended; determining a third search recommendation result based on at least one second object to be recommended.

[0145] It should be noted that this embodiment only describes a preferred embodiment of the present application. The second link may further include other multiple second task nodes, such as including a second recall node, a second screening node, and a second fine sorting node, etc. The embodiments of the present application do not limit the manner in which the above second task nodes form the second link. In addition, for the case where the second link includes a second recall node, a second screening node, and a second sorting node, the present application will be described in detail in subsequent embodiments and will not be elaborated here.

[0146] In this embodiment, when there is no pre-determined screening threshold for the multi-dimensional search related information in any dimension, the server processes the multi-dimensional search related information through the second link to obtain the second search recommendation result. This not only ensures the reliability of the search recommendation result, but also pre-determines the screening threshold corresponding to the search related information in at least one dimension based on the current search recommendation process, which helps to implement the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism in the subsequent process.

[0147] The above search recommendation method obtains the multi-dimensional search related information of the user. When there is a pre-determined screening threshold for the search related information in each dimension, it starts the first link, processes the multi-dimensional search related information through the first link to obtain the first search recommendation result; when there is no pre-determined screening threshold for the search related information in any dimension, it starts the second link, processes the multi-dimensional search related information through the second link to obtain the third search recommendation result. This not only ensures the reliability of the search recommendation result, but also pre-determines the screening threshold corresponding to the search related information in at least one dimension based on the current search recommendation process, which helps to implement the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism in the subsequent process, so as to shorten the total duration of the search recommendation and improve the response speed of the search recommendation.

[0148] In one embodiment, as Figure 5 shown, another search recommendation method is provided. Taking the method applied to the Figure 1 server 21 as an example, it includes the following steps:

[0149] S501: Obtain the multi-dimensional search related information of the user.

[0150] Specifically, S501 is the same as S401 and will not be elaborated here.

[0151] S502: Determine whether there is a pre-determined screening threshold for the search related information in each dimension. If yes, execute S503; if no, execute S509.

[0152] Specifically, S502 is the same as S402 and will not be elaborated here.

[0153] S503: Start the first link according to the multi-dimensional search related information.

[0154] Among them, the first link includes a first recall node, a first screening node, a first fine-ranking node, and a merging node arranged in sequence. Each first task node is used to execute different stage tasks, and data is transmitted concurrently and streamingly between each first task node.

[0155] Optionally, after the server 21 obtains the multi-dimensional search related information of the user through query processing, it sends the multi-dimensional search related information to the entry node of the first link (the first recall node in this embodiment), thereby starting the first link.

[0156] S504: The first recall node recalls at least one first candidate object according to the multi-dimensional search related information, and after recalling any first candidate object, transmits any first candidate object to the first screening node.

[0157] Optionally, the server 21 processes the multi-dimensional search related information such as the search keyword information, the geographical location information at the time of search, and the timestamp information at the time of search through the first recall node in the first link, recalls at least one first candidate object (such as a doc document), and after recalling any first candidate object, transmits any first candidate object to the first screening node.

[0158] Specifically, a recall engine is deployed on the first recall node. According to the preset recall policy, the engine combines the multi-dimensional search related information such as the search keyword information, the geographical location information at the time of search, and the timestamp information at the time of search, and recalls at least one first candidate object through at least one recall path. Each time a first candidate object is recalled, it is transmitted to the first screening node for further screening processing.

[0159] S505: The first screening node screens each incoming first candidate object to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmits any first object to be recommended to the first fine-ranking node.

[0160] Optionally, the server 21 combines the screening thresholds corresponding to the multi-dimensional search related information through the first screening node in the first link, screens each incoming first candidate object to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmits any first object to be recommended to the first fine-ranking node.

[0161] Specifically, the first screening node is deployed with a first rough ranking scoring model and a dynamic threshold model, and the dynamic threshold model caches the screening thresholds corresponding to the search-related information in each dimension. The screening threshold can be determined in advance based on the rough ranking scoring results and screening results of the data items in the second link. The server 21 performs rough ranking scoring on each incoming first candidate object through the first rough ranking scoring model in the first screening node to obtain the rough ranking scoring results corresponding to each first candidate object, and then through the dynamic threshold model in the first screening node, according to the rough ranking scoring results corresponding to each incoming first candidate object and the screening thresholds corresponding to the search-related information in each dimension, screens each incoming first candidate object to obtain at least one first object to be recommended. Each time the first screening node screens out a first object to be recommended, it transmits it to the first fine ranking node for further fine ranking processing.

[0162] In one embodiment, the first screening node further includes a deduplication table for recording the incoming first candidate objects; when screening each incoming first candidate object through the dynamic threshold model according to the rough ranking scoring results corresponding to each first candidate object and the screening thresholds corresponding to the search-related information in each dimension to obtain at least one first object to be recommended, and before transmitting any first object to be recommended to the first fine ranking node after obtaining it, the method further includes: determining whether the currently incoming first candidate object has been recorded in the deduplication table; if so, discarding the currently incoming first candidate object; if not, retaining the currently incoming first candidate object.

[0163] Among them, the deduplication table can be but is not limited to a data structure table such as a hash table for recording the incoming first candidate objects. By maintaining such a table, the server 21 can ensure that each first candidate object is screened and processed only once, even if the first candidate object may be recalled multiple times through multiple recall paths.

[0164] Optionally, after each first candidate object flows into the first screening node, the server 21 first determines whether the currently incoming first candidate object has been recorded in the deduplication table through the deduplication table. If the currently incoming first candidate object has been recorded in the deduplication table, the server 21 discards the currently incoming first candidate object, that is, the first candidate object no longer participates in the subsequent rough ranking scoring, dynamic threshold screening, fine ranking and other task stages; if the currently incoming first candidate object has not been recorded in the deduplication table, the server 21 retains the currently incoming first candidate object, that is, the first candidate object normally participates in the subsequent rough ranking scoring, dynamic threshold screening, fine ranking and other task stages.

[0165] It should be noted that this embodiment only describes a preferred implementation manner of the present application. After each first candidate object flows into the first screening node, the server 21 can also first perform rough ranking scoring and dynamic threshold screening on it, and then determine whether it has been recorded in the deduplication table. However, it is necessary to ensure that deduplication is performed before the data item enters the fine ranking to avoid prolonging the processing time in the fine ranking stage.

[0166] In this embodiment, by maintaining a deduplication table at the first screening node, the server can ensure that each first candidate object is only screened and processed once. Even if the first candidate object may be recalled multiple times through multiple recall paths, it avoids repeated processing of the same first candidate object, saves computing resources, and improves the computing speed. It also prevents the subsequent recommendation of the same content to the user and improves the user experience. The entire process is based on the deduplication mechanism, which can effectively save computing resources, shorten the total duration of search and recommendation, and improve the response speed of search and recommendation and the user experience.

[0167] S506: Fine-rank and score each first object to be recommended that flows in through the first fine-rank node to obtain a fine-rank scoring result corresponding to at least one first object to be recommended. After obtaining the fine-rank scoring result corresponding to any first object to be recommended, transmit any first object to be recommended and its corresponding fine-rank scoring result to the merging node.

[0168] Optionally, the first fine-rank node includes at least one fine-rank scoring model. The server 21 fine-ranks and scores each first object to be recommended that flows in through at least one fine-rank scoring model deployed in the first fine-rank node to obtain a fine-rank scoring result corresponding to at least one first object to be recommended. Each time the first fine-rank node obtains a fine-rank scoring result corresponding to a first object to be recommended, it transmits it to the merging node.

[0169] It can be understood that if multiple fine-rank scoring models are sequentially deployed in the first fine-rank node, the scoring result output by the previous fine-rank scoring model may be used as one of the input parameters of the subsequent fine-rank scoring model, thus forming a process of gradual evaluation. For example, the first fine-rank scoring model may perform fine-rank scoring based on scores and distances, and the second fine-rank scoring model adjusts it in combination with the user's historical preferences on this basis. The fine-rank scoring result corresponding to each first object to be recommended is output by the last fine-rank scoring model.

[0170] S507: Collect each first object to be recommended that flows in within the first preset time period and its corresponding fine-rank scoring result through the merging node.

[0171] Optionally, the server 21 collects each first object to be recommended flowing in within a first preset time period and its corresponding fine ranking score result through the merging node, thereby controlling the total time consumption of the entire first link. The start time of the first preset time period is determined by the start time of the first link. The duration of the first preset time period can be preset. For example, it can be set to 50 ms, 100 ms; it can also be adjusted according to actual needs, such as adjusted to 30 ms, 200 ms, etc.

[0172] In one embodiment, the method further includes: determining whether the merging node has collected a preset number of first objects to be recommended within a second preset time period; if the merging node has not collected a preset number of first objects to be recommended within the second preset time period, starting a backup link; processing the multi-dimensional search related information through the backup link according to a preset fallback strategy to obtain a second search recommendation result.

[0173] Among them, the start time of the second preset time period is determined by the start time of the first link. The duration of the second preset time period can be preset. The set duration of the second preset time period can be equal to the set duration of the first preset time period, or can be greater than or less than the set duration of the first preset time period. The preset number can be zero, or can be adjusted to a positive integer value according to actual needs. If the merging node has not collected a preset number of first objects to be recommended within the second preset time period, the server 21 determines that the first link is abnormal. The backup link refers to a data processing path started as an alternative when the main link (the first link and the second link in this embodiment) is abnormal. The preset fallback strategy is a set of preset rules or methods for dealing with data processing requirements in abnormal situations.

[0174] Exemplarily, if the server 21 detects that the merging node has not collected any first objects to be recommended within the second preset time period, it determines that the first link is abnormal, starts the backup link, and through the backup link, concurrently calls a simple recall path (such as an original word recall path) for fallback to obtain a second search recommendation result.

[0175] In this embodiment, the server starts the backup link by detecting the data items collected by the merging node when the merging node has not collected a preset number of first objects to be recommended; processes the multi-dimensional search related information through the backup link according to a preset fallback strategy to obtain a second search recommendation result, which can effectively handle the situation where the link is abnormal and improves the stability and reliability of the search recommendation.

[0176] S508: Determine a first search recommendation result based on each first object to be recommended collected by the merging node and its corresponding fine ranking score result.

[0177] Optionally, the server 21 uses at least one first object to be recommended with a relatively high fine ranking score among the first objects to be recommended collected by the merging node as at least one search recommendation object, and obtains a first search recommendation result.

[0178] Exemplarily, it is assumed that multiple fine ranking models are sequentially deployed on the first fine ranking node. Then, the fine ranking scores carried by each first object to be recommended collected by the merging node within the first preset time period are the fine ranking scores output by the last fine ranking model. The server 21 may sort all the first objects to be recommended collected by the merging node in descending order according to the fine ranking scores carried by each first object to be recommended, and then select the top N first objects to be recommended with relatively high fine ranking scores among all the first objects to be recommended collected as at least one search recommendation object, and use the at least one search recommendation object arranged in descending order of the fine ranking scores as the first search recommendation result.

[0179] It can be understood that after obtaining the first search recommendation result, the server 21 may directly send the first search recommendation result to the user terminal; or may continue to further process the first search recommendation result (such as re-ranking, etc.) and then send it to the user terminal. The embodiments of the present application do not make any limitations in this regard.

[0180] In this embodiment, by presetting the duration of the first preset time period, it can be ensured that within the first preset time period when the first link is started without any abnormalities, the server 21 can return a certain number of data items based on the results collected by the merging node. This not only helps to control the total time consumption of the entire first link, but also helps to effectively avoid the long-tail problem in performance (that is, the problem that the response time of the entire system is extended due to some requests not being completed for a long time) by restricting the maximum waiting time for data transmission. The entire process shortens the total duration of search recommendation and improves the response speed of search recommendation.

[0181] In one embodiment, after determining the first search recommendation result based on each first object to be recommended collected by the merging node and its corresponding fine ranking score, the method further includes: re-ranking the first search recommendation result to obtain a re-ranked first search recommendation result.

[0182] Optionally, after the server 21 collects each first object to be recommended flowing in within the preset time period and its corresponding fine ranking score through the merging node, it uses the at least one search recommendation object arranged in descending order of the fine ranking score as the first search recommendation result, and sends the first search recommendation result to the re-ranking node. The re-ranking node re-ranks the first search recommendation result through a preset diversity strategy to obtain a re-ranked first search recommendation result, and finally sends the re-ranked first search recommendation result to the user terminal.

[0183] In this embodiment, after screening and fine-ranking the recalled data items, the server further rearranges them. By introducing a diversity strategy, it can avoid the over-simplification of recommended content and improve the richness of search and recommended content and the user experience.

[0184] S509: Start the second link according to the multi-dimensional search related information.

[0185] Among them, the second link includes a second recall node, a second screening node, and a second fine-ranking node arranged in sequence. Each second task node is used to execute different stage tasks, and data is transmitted sequentially between each second task node.

[0186] Optionally, after the server 21 obtains the multi-dimensional search related information of the user through query processing, it sends the multi-dimensional search related information to the entry node of the second link (the second recall node in this embodiment), thereby starting the second link.

[0187] S510: The second recall node recalls at least one second candidate object according to the multi-dimensional search related information. After recalling at least one second candidate object, it transmits at least one second candidate object to the second screening node.

[0188] Optionally, the server 21 processes the multi-dimensional search related information such as search keyword information, geographical location information at the time of search, and timestamp information at the time of search through the second recall node in the second link, recalls at least one second candidate object (such as a doc document) through at least one recall path, and after recalling all the second candidate objects returned by all recall paths, transmits all the recalled second candidate objects to the second screening node.

[0189] Specifically, a recall engine is deployed on the second recall node. This engine can, according to a preset recall strategy (such as each dimension of search related information corresponding to a recall path), combine the multi-dimensional search related information such as search keyword information, geographical location information at the time of search, and timestamp information at the time of search, recall at least one second candidate object through at least one recall path, and after recalling all the second candidate objects, transmit all the recalled second candidate objects to the second screening node for further screening processing.

[0190] S511: The second screening node screens at least one second candidate object to obtain at least one second object to be recommended. After obtaining at least one second object to be recommended, it transmits at least one second object to be recommended to the second fine-ranking node.

[0191] Optionally, the server 21 screens each of the at least one second candidate object recalled through a second screening node in the second link to obtain at least one second object to be recommended. After all the second objects to be recommended are screened out, all the screened second objects to be recommended are transmitted to the second fine ranking node.

[0192] Specifically, the second screening node is deployed with a second rough ranking scoring model. The server 21 respectively performs rough ranking scoring on all the recalled second candidate objects through the second rough ranking scoring model to obtain the rough ranking scoring results corresponding to each second candidate object, and takes at least one second candidate object with a higher rough ranking scoring result among all the recalled second candidate objects as at least one second object to be recommended. After all the second objects to be recommended are obtained, all the second objects to be recommended are transmitted to the second fine ranking node for further fine ranking processing.

[0193] In one embodiment, the screening threshold is determined by the rough ranking scoring results corresponding to at least one second candidate object.

[0194] It can be understood that each recall path can correspond to one-dimensional search related information. For example, the recall engine of the second recall node can recall one-way data respectively according to various dimensions of search related information such as the search keyword information obtained by query processing and the geographical location information at the time of search. The screening thresholds corresponding to the various dimensions of search related information can be determined by the rough ranking scoring results corresponding to at least one second candidate object in each recall path.

[0195] Optionally, the server 21 can sort the second candidate objects in all the recall paths in descending order according to the rough ranking scoring results output by the rough ranking scoring model, and then select the top N second objects to be recommended with higher rough ranking scoring results from the second candidate objects in each recall path as at least one search recommendation object according to the sorting result, and take the lowest value of the rough ranking scoring results among the top N second objects to be recommended corresponding to each recall path as the screening threshold corresponding to each dimension of search related information.

[0196] In this embodiment, each dimension of search related information is used to recall each path of second candidate objects, and the screening threshold corresponding to each dimension of search related information is determined according to the rough ranking scoring results corresponding to the second candidate objects recalled in each path, which can not only ensure the reliability of the search recommendation results, but also pre-determine the screening threshold corresponding to at least one dimension of search related information based on the current search recommendation process, which helps to implement the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism in the subsequent process, so as to shorten the total duration of search recommendation and improve the response speed of search recommendation.

[0197] S512: The second fine-ranking node performs fine-ranking scoring on at least one second object to be recommended, and obtains the fine-ranking scoring results corresponding to each second object to be recommended.

[0198] Optionally, the second fine-ranking node includes at least one fine-ranking scoring model. The server 21 performs fine-ranking scoring on all the filtered second objects to be recommended respectively through the at least one fine-ranking scoring model deployed in the second fine-ranking node, and obtains the fine-ranking scoring results corresponding to each of all the second objects to be recommended.

[0199] It can be understood that if multiple fine-ranking scoring models are successively deployed in the second fine-ranking node, the scoring results output by the previous fine-ranking scoring model may be used as one of the input parameters of the subsequent fine-ranking scoring model, thus forming a process of gradual evaluation. For example, the first fine-ranking scoring model may perform fine-ranking scoring based on score and distance, and the second fine-ranking scoring model adjusts based on the user's historical preferences on this basis. The fine-ranking scoring results corresponding to each second object to be recommended are output by the last fine-ranking scoring model.

[0200] S513: Based on at least one second object to be recommended and its corresponding fine-ranking scoring result, determine the third search recommendation result.

[0201] Optionally, after the server 21 obtains the fine-ranking scoring results corresponding to all the second objects to be recommended through the second fine-ranking node, it determines the third search recommendation result based on all the second objects to be recommended and their corresponding fine-ranking scoring results.

[0202] Exemplarily, assume that multiple fine-ranking scoring models are successively deployed in the second fine-ranking node. Then the server 21 may sort all the second objects to be recommended in descending order according to the fine-ranking scoring results output by the last fine-ranking scoring model, and then select the top N second objects to be recommended with higher fine-ranking scoring results among all the second objects to be recommended as at least one search recommendation object, and use the at least one search recommendation object arranged in descending order of fine-ranking scoring results as the third search recommendation result.

[0203] It can be understood that after the server 21 obtains the third search recommendation result, it may directly send the third search recommendation result to the user terminal; or it may continue to further process the third search recommendation result (such as re-ranking, etc.) and then send it to the user terminal. The embodiments of the present application do not make any limitations on this.

[0204] In this embodiment, by presetting the duration of the first preset time period, it is possible to ensure that within the first preset time period of starting the first link without anomalies, the server 21 can return a certain number of data items based on the results collected by the merging node. This not only helps to control the total time consumption of the entire first link, but also effectively avoids the long-tail problem in performance (i.e., the problem that the response time of the entire system is extended due to some requests not being completed for a long time) by restricting the maximum waiting time for data transmission. The entire process shortens the total duration of search and recommendation and improves the response speed of search and recommendation.

[0205] The above search and recommendation method obtains multi-dimensional search-related information of a user. When there are predetermined screening thresholds for each dimension of the search-related information, the first link is started, and the multi-dimensional search-related information is processed through the first link to obtain a first search and recommendation result. When there is no predetermined screening threshold for any dimension of the search-related information, the second link is started, and the multi-dimensional search-related information is processed through the second link to obtain a third search and recommendation result. This can not only ensure the reliability of the search and recommendation results, but also determine at least one dimension of the screening threshold corresponding to the search-related information based on the current search and recommendation process, which helps to realize the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism in the future, so as to shorten the total duration of search and recommendation and improve the response speed of search and recommendation.

[0206] In one embodiment, as Figure 6 shown, another search and recommendation method is provided. Taking the method applied to Figure 1 the server 21 in

[0207] S601: Obtain multi-dimensional search-related information of the user.

[0208] Specifically, S601 is the same as S401 and will not be elaborated here.

[0209] S602: Determine whether there are predetermined screening thresholds for each dimension of the search-related information. If so, execute S603; if not, execute S613.

[0210] Specifically, S602 is the same as S402 and will not be elaborated here.

[0211] S603: Start the first link according to the multi-dimensional search-related information.

[0212] Specifically, S603 is the same as S503 and will not be elaborated here.

[0213] S604: Recall at least one first candidate object according to the multi-dimensional search related information by the first recall node, and after recalling any first candidate object, transmit any first candidate object to the first screening node.

[0214] Specifically, S604 is the same as S504 and will not be elaborated here.

[0215] S605: Coarsely rank and score each incoming first candidate object through the first coarse ranking scoring model to obtain the coarse ranking scoring results corresponding to each first candidate object.

[0216] It can be understood that the first screening node is deployed with a first coarse ranking scoring model and a dynamic threshold model, and the dynamic threshold model caches the screening thresholds corresponding to the multi-dimensional search related information. The screening threshold can be determined in advance based on the coarse ranking scoring results and screening results of the data items in the second link. Optionally, the server 21 coarsely ranks and scores each incoming first candidate object through the first coarse ranking scoring model in the first screening node to obtain the coarse ranking scoring results corresponding to each first candidate object, and then concurrently and streamingly transmits the coarse ranking scoring results corresponding to each first candidate object to the dynamic threshold model.

[0217] S606: Screen each incoming first candidate object through the dynamic threshold model according to the coarse ranking scoring results corresponding to each incoming first candidate object and the screening thresholds corresponding to the multi-dimensional search related information to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmit any first object to be recommended to the first fine ranking node.

[0218] Among them, the dynamic threshold model pre-caches the screening thresholds corresponding to the multi-dimensional search related information.

[0219] Optionally, the server 21 screens each incoming first candidate object through the dynamic threshold model in the first screening node according to the coarse ranking scoring results corresponding to each incoming first candidate object and the screening thresholds corresponding to the multi-dimensional search related information to obtain at least one first object to be recommended. Every time the first screening node screens out a first object to be recommended, it transmits it to the first fine ranking node for further fine ranking processing.

[0220] In this embodiment, by deploying a coarse ranking scoring model and a dynamic threshold model in the first screening node, the server can reasonably screen the first candidate objects in combination with information such as the user's search term and the geographical location when the user searches, which can effectively improve the rationality and accuracy of the search recommendation. While shortening the total duration of the search recommendation and improving the response speed of the search recommendation, the quality of the search recommendation is improved.

[0221] S607: The first fine-ranking node performs fine-ranking scoring on each incoming first object to be recommended, obtains the fine-ranking scoring results corresponding to at least one first object to be recommended, and after obtaining the fine-ranking scoring result corresponding to any first object to be recommended, transmits any first object to be recommended and its corresponding fine-ranking scoring result to the merging node.

[0222] Specifically, S607 is the same as S506 and will not be elaborated here.

[0223] S608: The merging node collects each incoming first object to be recommended and its corresponding fine-ranking scoring result within the first preset time period.

[0224] Specifically, S608 is the same as S507 and will not be elaborated here.

[0225] S609: Determine whether the first objects to be recommended collected by the merging node include the specified objects to be recommended. If yes, execute S611; if not, execute S610.

[0226] Among them, the specified objects to be recommended are recalled by a pre-specified key recall path.

[0227] Exemplarily, each data item recalled by the original word recall path is more in line with the user's current needs and has a higher conversion rate. Therefore, each object to be recommended recalled by the original word recall path can be used as the specified object to be recommended. The server 20 can determine whether the merging node has collected each object to be recommended recalled by the original word recall path at the end of the first preset time period.

[0228] If the merging node has collected the specified objects to be recommended, at least one first object to be recommended with a higher fine-ranking scoring result among the first objects to be recommended collected by the merging node and at least one specified object to be recommended with a higher fine-ranking scoring result among the specified objects to be recommended collected by the merging node can be used as at least one search recommendation object to obtain the first search recommendation result; if the merging node has not collected the specified objects to be recommended yet, after the first preset time period, continue to collect the specified objects to be recommended through the merging node until the first objects to be recommended collected by the merging node include the specified objects to be recommended, so as to ensure the quality of search recommendations.

[0229] S610: The merging node collects the specified objects to be recommended and executes S609 again.

[0230] Optionally, if the server 21 determines that the first objects to be recommended collected by the merging node do not include the specified objects to be recommended (such as each object to be recommended recalled by the original word recall path), after exceeding the first preset time period, continue to collect the specified objects to be recommended through the merging node until the first objects to be recommended collected by the merging node include the specified objects to be recommended, so as to ensure the quality of search recommendations.

[0231] S611: Use at least one first candidate object with a relatively high fine ranking score among the first candidate objects collected by the merging node, and at least one specified candidate object with a relatively high fine ranking score among the specified candidate objects collected by the merging node, as at least one search recommendation object, to obtain a first search recommendation result.

[0232] Optionally, if the server 21 determines that the first candidate objects collected by the merging node already include the specified candidate objects (such as each candidate object recalled by the original word recall path), then at least one first candidate object with a relatively high fine ranking score among the first candidate objects collected by the merging node, and at least one specified candidate object with a relatively high fine ranking score among the specified candidate objects collected by the merging node, can be used as at least one search recommendation object, to obtain a first search recommendation result.

[0233] In this embodiment, the server determines whether the first candidate objects collected by the merging node include the specified candidate objects. When the merging node has not collected the specified candidate objects, the server continues to collect the specified candidate objects through the merging node until the first candidate objects collected by the merging node include the specified candidate objects; when the merging node has collected the specified candidate objects, at least one first candidate object with a relatively high fine ranking score among the first candidate objects collected by the merging node, and at least one specified candidate object with a relatively high fine ranking score among the specified candidate objects collected by the merging node, are used as at least one search recommendation object, to obtain a first search recommendation result. This can ensure that the existing search recommendation results are returned to the user terminal while guaranteeing the quality of search recommendations.

[0234] S612: Re-rank the first search recommendation result to obtain a re-ranked first search recommendation result.

[0235] Optionally, after the server 21 collects each first candidate object flowing in within a preset time period and its corresponding fine ranking score through the merging node, at least one search recommendation object ranked from high to low according to the fine ranking score is used as the first search recommendation result, and this first search recommendation result is sent to the re-ranking node. The re-ranking node re-ranks the first search recommendation result through a preset diversity strategy to obtain a re-ranked first search recommendation result, and finally sends the re-ranked first search recommendation result to the user terminal.

[0236] In this embodiment, after screening and fine ranking the recalled data items, the server further re-ranks them. By introducing a diversity strategy, it can avoid the recommended content from being too single, and improve the richness of search recommendation content and the user experience.

[0237] S613: Initiate the second link according to the multi-dimensional search related information.

[0238] Specifically, S613 is the same as S509, which will not be elaborated here.

[0239] S614: Recall at least one second candidate object by the second recall node according to the multi-dimensional search related information, and after recalling at least one second candidate object, transmit at least one second candidate object to the second screening node.

[0240] Specifically, S615 is the same as S510, which will not be elaborated here.

[0241] S615: Coarsely rank and score at least one second candidate object respectively through the second coarse ranking scoring model to obtain the coarse ranking scoring results corresponding to each second candidate object.

[0242] Optionally, the server 21 coarsely ranks and scores all the recalled second candidate objects respectively through the second coarse ranking scoring model deployed on the second coarse ranking node to obtain the coarse ranking scoring results corresponding to each second candidate object.

[0243] S616: Take at least one second candidate object with a higher coarse ranking scoring result among at least one second candidate object as at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmit at least one second object to be recommended to the second fine ranking node.

[0244] Optionally, after obtaining the coarse ranking scoring results corresponding to all second candidate objects, the server 21 takes at least one second candidate object with a higher coarse ranking scoring result among all the recalled second candidate objects as at least one second object to be recommended, and after obtaining all second objects to be recommended, transmits all second objects to be recommended to the second fine ranking node for further fine ranking processing.

[0245] It should be noted that the screening thresholds corresponding to the multi-dimensional search related information can be determined by the coarse ranking scoring results output by the coarse ranking scoring model and the final screening results. It can be understood that each recall path can correspond to one-dimensional search related information. For example, the recall engine of the second recall node can recall one path of data respectively according to various dimensions of search related information such as the search keyword information obtained by query processing and the geographical location information during the search, and the screening thresholds corresponding to the multi-dimensional search related information can be determined by the coarse ranking scoring results corresponding to at least one second candidate object in each recall path.

[0246] Exemplarily, the server 21 may, according to the rough ranking score results output by the rough ranking score model, sort the second candidate objects in all recall paths in descending order of the score results, and then, according to the sorting results, select the top N second objects to be recommended with higher rough ranking score results from the second candidate objects in each recall path as at least one search recommendation object, and use the lowest value of the rough ranking score results among the top N second objects to be recommended corresponding to each recall path as the screening threshold corresponding to each dimension of search-related information.

[0247] In this embodiment, by recalling the second candidate objects in each path through each dimension of search-related information and determining the screening threshold corresponding to each dimension of search-related information according to the rough ranking score results corresponding to the second candidate objects recalled in each path, it can not only ensure the reliability of the search recommendation results, but also pre-determine the screening threshold corresponding to at least one dimension of search-related information based on the current search recommendation process, which helps to subsequently implement the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism, so as to shorten the total duration of the search recommendation and improve the response speed of the search recommendation.

[0248] S617: The second fine-ranking node performs fine-ranking scoring on at least one second object to be recommended to obtain the fine-ranking score results corresponding to each second object to be recommended.

[0249] Specifically, S618 is the same as S512 and will not be elaborated here.

[0250] S618: Based on at least one second object to be recommended and its corresponding fine-ranking score results, determine the third search recommendation result.

[0251] Specifically, S619 is the same as S513 and will not be elaborated here.

[0252] S619: Re-rank the third search recommendation result to obtain the re-ranked third search recommendation result.

[0253] Optionally, after the server 21 obtains the third search recommendation result through the second fine-ranking node, it may continue to send the third search recommendation result to the re-ranking node, re-rank the third search recommendation result through the diversity strategy preset by the re-ranking node to obtain the re-ranked third search recommendation result, and finally send the re-ranked third search recommendation result to the user terminal.

[0254] The above search recommendation method obtains multi-dimensional search-related information of a user. When there are predetermined screening thresholds for the search-related information in each dimension, the first link is activated, and the multi-dimensional search-related information is processed through the first link to obtain a first search recommendation result. When there is no predetermined screening threshold for any dimension of the search-related information, the second link is activated, and the multi-dimensional search-related information is processed through the second link to obtain a third search recommendation result. This not only ensures the reliability of the search recommendation result but also determines in advance the screening thresholds corresponding to the search-related information in at least one dimension based on the current search recommendation process, which helps to implement the concurrent processing of tasks in different stages and the continuous transmission of data based on the concurrent streaming execution mechanism in the subsequent process, so as to shorten the total duration of the search recommendation and improve the response speed of the search recommendation.

[0255] It should be understood that although the steps in the flowcharts involved in the above embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0256] Based on the inventive concept of the above search recommendation method, as Figure 7 shown, an embodiment of the present application further provides a search recommendation device 700 for implementing the above-mentioned search recommendation method. The search recommendation device 700 includes:

[0257] An acquisition module 701, configured to acquire multi-dimensional search-related information of a user;

[0258] A start module 702, configured to activate the first link according to the multi-dimensional search-related information; the first link includes a plurality of first task nodes, and each first task node is used to execute different stage tasks, and data is transmitted concurrently and in a streaming manner between the first task nodes;

[0259] A processing module 703, configured to process the multi-dimensional search-related information through the first link to obtain a first search recommendation result.

[0260] In one embodiment, the multiple first task nodes include at least one of a first recall node, a first screening node, and a first ranking node, and a merging node; the first ranking node includes a first fine ranking node; the merging node is configured to collect the data flowing in within a first preset time period; the start time of the first preset time period is determined by the start time of the first link.

[0261] In one embodiment, the multiple first task nodes include a first screening node and a merging node; the merging node is configured to collect the data flowing in within a first preset time period; the start time of the first preset time period is determined by the start time of the first link; the search and recommendation device 700 further includes a judgment module, configured to judge whether there is a preset screening threshold for each dimension of search-related information; if there is a preset screening value for each dimension of search-related information, then execute the step of starting the first link according to the multi-dimensional search-related information.

[0262] In one embodiment, the multiple first task nodes further include a first recall node and a first fine ranking node; the processing module 703 is specifically configured to recall at least one first candidate object according to the multi-dimensional search-related information through the first recall node, and after recalling any first candidate object, transmit any first candidate object to the first screening node; screen each first candidate object flowing in through the first screening node to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmit any first object to be recommended to the first fine ranking node; perform fine ranking scoring on each first object to be recommended flowing in through the first fine ranking node to obtain a fine ranking scoring result corresponding to at least one first object to be recommended, and after obtaining a fine ranking scoring result corresponding to any first object to be recommended, transmit any first object to be recommended and its corresponding fine ranking scoring result to the merging node; collect each first object to be recommended flowing in within a first preset time period and its corresponding fine ranking scoring result through the merging node; determine a first search and recommendation result based on each first object to be recommended collected by the merging node and its corresponding fine ranking scoring result.

[0263] In one embodiment, the first screening node includes a first rough ranking scoring model and a dynamic threshold model; the processing module 703 is specifically configured to perform rough ranking scoring on each first candidate object flowing in through the first rough ranking scoring model to obtain a rough ranking scoring result corresponding to each first candidate object; screen each first candidate object flowing in through the dynamic threshold model according to the rough ranking scoring result corresponding to each first candidate object flowing in and the screening threshold corresponding to each dimension of search-related information to obtain at least one first object to be recommended, and after obtaining any first object to be recommended, transmit any first object to be recommended to the first fine ranking node.

[0264] In one embodiment, the first screening node further includes a deduplication table for recording the first candidate objects that have flowed in; the above-mentioned judgment module is further configured to judge whether the currently flowing-in first candidate object has been recorded in the deduplication table; if so, discard the currently flowing-in first candidate object; if not, retain the currently flowing-in first candidate object.

[0265] In one embodiment, the above-mentioned judgment module is further configured to judge whether the merging node has collected a preset number of first objects to be recommended within a second preset time period; if the merging node has not collected a preset number of first objects to be recommended within the second preset time period, start a backup link; process the multi-dimensional search related information according to a preset fallback strategy through the backup link to obtain a second search recommendation result.

[0266] In one embodiment, the processing module 703 is specifically configured to use at least one first object to be recommended with a higher refined ranking score among the first objects to be recommended collected by the merging node as at least one search recommendation object to obtain a first search recommendation result.

[0267] In one embodiment, the processing module 703 is further configured to, when the first objects to be recommended collected by the merging node do not include the specified objects to be recommended, collect the specified objects to be recommended through the merging node; the specified objects to be recommended are recalled by a pre-specified key recall path; use at least one first object to be recommended with a higher refined ranking score among the first objects to be recommended collected by the merging node, and at least one specified object to be recommended with a higher refined ranking score among the specified objects to be recommended collected by the merging node as at least one search recommendation object to obtain a first search recommendation result.

[0268] In one embodiment, the processing module 703 is further configured to re-rank the first search recommendation result to obtain a re-ranked first search recommendation result.

[0269] In one embodiment, the processing module 703 is further configured to, if there is no predetermined screening threshold for the multi-dimensional search related information, start a second link according to the multi-dimensional search related information; the second link includes a plurality of second task nodes, and each second task node is used to execute different stage tasks, and data is transmitted sequentially between the second task nodes; process the multi-dimensional search related information through the second link to obtain a third search recommendation result.

[0270] In one embodiment, the plurality of second task nodes include at least one of a second recall node and a second sorting node, and a second screening node; the second sorting node includes a second refined ranking node.

[0271] In one embodiment, the multiple second task nodes include a second recall node, a second screening node, and a second fine ranking node; the processing module 703 is further specifically configured to recall at least one second candidate object by the second recall node according to the multi-dimensional search related information, and after recalling at least one second candidate object, transmit at least one second candidate object to the second screening node; screen at least one second candidate object by the second screening node to obtain at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmit at least one second object to be recommended to the second fine ranking node; perform fine ranking scoring on at least one second object to be recommended by the second fine ranking node to obtain the fine ranking scoring results corresponding to each second object to be recommended; and determine the third search recommendation result based on at least one second object to be recommended and its corresponding fine ranking scoring results.

[0272] In one embodiment, the second screening node includes a second rough ranking scoring model; the processing module 703 is further specifically configured to perform rough ranking scoring on at least one second candidate object respectively by the second rough ranking scoring model to obtain the rough ranking scoring results corresponding to each second candidate object; use at least one second candidate object with a higher rough ranking scoring result among at least one second candidate object as at least one second object to be recommended, and after obtaining at least one second object to be recommended, transmit at least one second object to be recommended to the second fine ranking node.

[0273] In one embodiment, the screening threshold is determined by the rough ranking scoring results corresponding to at least one second candidate object.

[0274] Please refer to Figure 8 , the embodiment of the present application further provides a search recommendation device 800 for implementing the above search recommendation method applied to a user terminal. The search recommendation device 800 includes:

[0275] A first receiving module 801, configured to receive a search request of a user;

[0276] A sending module 802, configured to send the search request to a server, so that the server, in response to the search request, obtains multi-dimensional search related information of the user according to the search information carried in the search request, starts a first link according to the multi-dimensional search related information, processes the multi-dimensional search related information through the first link to obtain a first search recommendation result, and sends the first search recommendation result to the user terminal; wherein, the first link includes multiple first task nodes, each first task node is used to execute different stage tasks, and data is transmitted between each first task node in a concurrent streaming manner;

[0277] A second receiving module 803, configured to receive the first search recommendation result;

[0278] A display module 804, configured to perform display based on the first search recommendation result.

[0279] Each module in the above-mentioned search and recommendation device 700 and search and recommendation device 800 can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0280] The embodiment of the present application also provides an electronic device, which can be a server, and its internal structure diagram can be as Figure 9 shown. The electronic device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store search data sources of the online search platform, etc. The input / output interface of the electronic device is used to exchange information between the processor and external devices. The communication interface of the electronic device is used to communicate with external terminals through a network connection. The processor of the electronic device executes a computer program to implement a search and recommendation method.

[0281] Those skilled in the art can understand that Figure 9 the structure shown in

[0282] It should be noted that the electronic device can also be a terminal, and its internal structure diagram can include a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the electronic device is used to exchange information between the processor and external devices. The communication interface of the electronic device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a search and recommendation method. The display unit of the electronic device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, touchpad, or mouse, etc.

[0283] In one embodiment, an electronic device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0284] The embodiment of the present application further provides a computer storage medium. Instructions are stored in the computer storage medium. When they run on a computer or a processor, the computer or the processor is enabled to execute one or more steps in the above embodiments. If the respective component modules of the above electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium.

[0285] In one embodiment, a computer storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0286] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer storage medium or transmitted through the computer storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a Digital Versatile Disc (DVD)), or a semiconductor medium (such as a Solid State Disk (SSD)), etc.

[0287] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disks, or optical discs. Without conflict, the technical features in this embodiment and the implementation solutions can be combined arbitrarily.

[0288] The above-described embodiments are merely described in a preferred implementation manner of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present application should fall within the protection scope determined by the claims.

[0289] The above description has been made of specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order shown or sequential order to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A search recommendation method, characterized in that: The method comprises: Obtain information related to the user's multi-dimensional search; The first link is started according to the multi-dimensional search related information; the first link includes a plurality of first task nodes, each of the first task nodes is used to perform a task at a different stage, and each of the first task nodes concurrently streams data; The multi-dimensional search related information is processed through the first link to obtain a first search recommendation result.

2. The method according to claim 1, characterized in that The multiple first task nodes include at least one of a first recall node, a first screening node, a first sorting node, and a merging node; the first sorting node includes a first fine sorting node; the merging node is used to collect data flowing in within a first preset time period; the starting time of the first preset time period is determined by the start time of the first link.

3. The method according to claim 1, characterized in that The plurality of first task nodes include a first screening node and a merging node; the merging node is used to collect data flowing in within a first preset time period; The starting time of the first preset time period is determined by the start time of the first link; Before initiating the first link according to the multi-dimensional search related information, the method further includes: Determining whether there is a predetermined screening threshold for the search-related information in each dimension; If there is a predetermined screening value for each dimension of the search related information, the step of starting the first link according to the multi-dimensional search related information is performed.

4. The method according to claim 3, characterized in that The method further comprises: If there is no predetermined screening threshold for the search-related information in each dimension, a second link is started according to the multi-dimensional search-related information; the second link includes a plurality of second task nodes, each of which is used to perform a task at a different stage, and each of the second task nodes transmits data in sequence; The multi-dimensional search related information is processed through the second link to obtain a third search recommendation result.

5. The method according to claim 4, characterized in that The plurality of second task nodes include at least one of a second recall node, a second sorting node, and a second screening node; the second sorting node includes a second fine sorting node.

6. The method according to claim 5, characterized in that The plurality of second task nodes include the second recall node, the second screening node and the second fine sorting node; The processing of the multi-dimensional search related information through the second link to obtain a third search recommendation result includes: Recalling at least one second candidate object according to the multi-dimensional search related information through the second recall node, and after recalling the at least one second candidate object, transmitting the at least one second candidate object to the second screening node; Screening the at least one second candidate object through the second screening node to obtain at least one second object to be recommended, and after obtaining the at least one second object to be recommended, transmitting the at least one second object to be recommended to the second refinement node; Performing fine ranking and scoring on the at least one second object to be recommended by the second fine ranking node to obtain a fine ranking and scoring result corresponding to each second object to be recommended; Based on the at least one second object to be recommended and its corresponding refined ranking score result, the third search recommendation result is determined.

7. A search recommendation method, characterized in that: Applied to a user terminal, the method comprises: Receive a user's search request; The search request is sent to a server, so that the server responds to the search request, obtains the multi-dimensional search related information of the user according to the search information carried in the search request, starts a first link according to the multi-dimensional search related information, processes the multi-dimensional search related information through the first link, obtains a first search recommendation result, and sends the first search recommendation result to the user terminal; wherein the first link includes a plurality of first task nodes, each of which is used to perform a different stage task, and each of the first task nodes concurrently streams data; receiving the first search recommendation result; Displaying recommended results based on the first search.

8. A search recommendation system, characterized in that: The search recommendation system comprises: a user terminal and a server; wherein, The merchant terminal is used to receive a user's search request and send the search request to a server; The server is configured to respond to the search request, obtain the multi-dimensional search related information of the user according to the search information carried in the search request, start a first link according to the multi-dimensional search related information, process the multi-dimensional search related information through the first link, obtain a first search recommendation result, and send the first search recommendation result to the user terminal; wherein the first link includes a plurality of first task nodes, each of which is used to perform a task at a different stage, and each of the first task nodes concurrently streams data; The merchant terminal is further used to receive the first search recommendation result and display it based on the first search recommendation result.

9. An electronic device, characterized in that: include: A processor and a memory; the memory stores a computer program, and the processor implements the method steps of any one of claims 1 to 7 when executing the computer program.

10. A computer storage medium, characterized in that: The computer storage medium stores a plurality of instructions, which are suitable for being loaded by a processor and executing the method steps according to any one of claims 1 to 7.

Images