CN118035511A - Recommendation platform, recommendation method, electronic equipment and storage medium - Google Patents

Abstract

The invention discloses a recommendation platform, a recommendation method, electronic equipment and a storage medium. The platform comprises: the request acquisition module is used for acquiring a recommendation request of a calling party; the recommendation module determining module is used for determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist; and the recommendation result determining module is used for determining a target recommendation result according to the recommendation request and the target recommendation module and returning the target recommendation result to the calling party. According to the embodiment of the invention, when the recommendation platform does not have the corresponding target recommendation result of the recommendation request of the calling party, the corresponding target recommendation module is determined according to the recommendation request, and the final target recommendation result is determined by the recommendation request and the target recommendation module, so that the feedback efficiency of the recommendation result is improved to a certain extent, the personalized recommendation process supporting each recommendation request is realized, the recommendation result is more accurate, and the use experience is further effectively improved.

Description

Recommendation platform, recommendation method, electronic equipment and storage medium

Technical Field

The present invention relates to the field of computer technologies, and in particular, to a recommendation platform, a recommendation method, an electronic device, and a storage medium.

Background

The intelligent recommendation platform based on the machine learning technology has become a necessary support for many Internet services through long-term development, and has wide application in the fields of IPTV, electronic commerce, search, social contact, resources and the like.

Some existing recommendation platforms or systems cannot support personalized recommendation flows, and recommendation effects are not ideal; meanwhile, large-scale data processing consumes more time, better real-time performance and accurate recommendation effect cannot be achieved, and the use experience is poor.

Disclosure of Invention

The invention provides a recommendation platform, a recommendation method, electronic equipment and a storage medium, which are used for determining a corresponding target recommendation module according to a recommendation request when the recommendation platform does not have the corresponding target recommendation result of the recommendation request of a calling party, and determining a final target recommendation result by the recommendation request and the target recommendation module, so that the feedback efficiency of the recommendation result is improved to a certain extent, the personalized recommendation flow supporting each recommendation request is realized, the recommendation result is more accurate, and the use experience is further effectively improved.

According to an aspect of the present invention, there is provided a recommendation platform comprising:

the request acquisition module is used for acquiring a recommendation request of a calling party;

the recommendation module determining module is used for determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist;

and the recommendation result determining module is used for determining a target recommendation result according to the recommendation request and the target recommendation module and returning the target recommendation result to the calling party.

According to another aspect of the present invention, there is provided a recommendation method applied to a recommendation platform, the recommendation method including:

Acquiring a recommendation request of a calling party;

When a target recommendation result corresponding to the recommendation request does not exist, determining a corresponding target recommendation module according to the recommendation request;

And determining a target recommendation result according to the recommendation request and the target recommendation module, and returning the target recommendation result to the calling party.

According to another aspect of the present invention, there is provided an electronic apparatus including:

At least one processor; and

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

The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the recommendation method according to any of the embodiments of the present invention.

According to another aspect of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to execute the recommendation method according to any of the embodiments of the present invention.

The embodiment of the invention provides a recommendation platform, which comprises the following components: the request acquisition module is used for acquiring a recommendation request of a calling party; the recommendation module determining module is used for determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist; and the recommendation result determining module is used for determining a target recommendation result according to the recommendation request and the target recommendation module and returning the target recommendation result to the calling party. According to the embodiment of the invention, when the recommendation platform does not have the corresponding target recommendation result of the recommendation request of the calling party, the corresponding target recommendation module is determined according to the recommendation request, and the final target recommendation result is determined by the recommendation request and the target recommendation module, so that the feedback efficiency of the recommendation result is improved to a certain extent, the personalized recommendation process supporting each recommendation request is realized, the recommendation result is more accurate, and the use experience is further effectively improved.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the invention or to delineate the scope of the invention. Other features of the present invention will become apparent from the description that follows.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a recommendation platform according to a first embodiment of the present invention;

FIG. 2 is a schematic diagram of a recommendation platform according to a second embodiment of the present invention;

FIG. 3 is a schematic diagram of a recommendation platform according to a third embodiment of the present invention;

FIG. 4 is a schematic diagram of a recommended business process according to a third embodiment of the present invention;

FIG. 5 is a schematic architecture diagram of a recommendation platform according to a third embodiment of the present invention;

FIG. 6 is a flowchart of a recommendation method according to a fourth embodiment of the present invention;

Fig. 7 is a schematic structural diagram of an electronic device implementing the recommendation method according to the embodiment of the present invention.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Example 1

Fig. 1 is a schematic diagram of a recommendation platform provided in a first embodiment of the present invention, where the embodiment is applicable to a case of responding to a recommendation request of a caller and returning a corresponding recommendation result in recommendation scenes such as commodity recommendation and movie and television sheet recommendation. As shown in fig. 1, a recommendation platform provided in the first embodiment includes: a request acquisition module 10, a recommendation module determining module 20 and a recommendation result determining module 30. The following describes the structural composition of the recommendation platform of the present embodiment in detail.

The request acquisition module 10 is configured to acquire a recommendation request of a caller.

The caller may refer to a user client for providing a recommendation request or other service platforms with recommendation requirements, and the caller may be a terminal capable of performing network communication, such as a mobile phone, a tablet computer, a desktop computer, or the like, or may be a service platform in communication connection with the recommendation platform in this embodiment, such as a music platform, a video platform, a shopping platform, or the like, which is not particularly limited in this embodiment of the present invention.

The recommendation request may be a request directed to a caller to recommend a desired recommendation result, for example, the recommendation request may be a request issued by a video platform to recommend a related movie and television sheet for user a; as another example, the recommendation request may also be a request issued by the shopping platform to recommend related items for user B, and so on.

In the embodiment of the present invention, the recommendation platform may acquire the recommendation request of the caller through the request acquisition module 10, and in practical application, the recommendation request may be generated by triggering the user according to the recommendation requirement of the user, or the recommendation request may be sent by the service platform with the recommendation requirement for the target user, which is not limited in particular in the embodiment of the present invention.

The recommendation module determining module 20 is configured to determine, when there is no target recommendation result corresponding to the recommendation request, a corresponding target recommendation module according to the recommendation request.

The target recommendation result may refer to a recommendation result provided by the recommendation platform and corresponding to the recommendation request. The target recommendation module may refer to a recommendation module included in an expected recommendation flow determined by a recommendation request, and the recommendation module may include various algorithms and modules related to recommendation flow services, and the like, and illustratively, the recommendation module may include recall modules such as collaborative filtering, matrix decomposition, vector recall, and the like, and sort modules such as coarse sort, fine sort, rearrangement, and the like.

In the embodiment of the present invention, after the recommendation platform receives the recommendation request of the caller, the recommendation module determining module 20 may query whether the cache of the recommendation platform already stores the corresponding target recommendation result according to the information such as the user identifier and the query identifier corresponding to the recommendation request, and if the query hits, directly returns the target recommendation result stored in the cache to the caller, without executing the subsequent recommendation flow; if the query is not hit, the recommendation module determining module 20 may parse the recommendation request to obtain an expected recommendation flow, and determine a target recommendation module corresponding to the recommendation request based on the expected recommendation flow. It should be noted that the expected recommendation process may be a character string expressing the expected recommendation process, where the character string includes field information of all required target recommendation modules.

It can be understood that after each recommendation process is executed by the recommendation platform, the corresponding recommendation result can be stored in the cache, and although the memory occupation of the whole recommendation platform can be improved to a certain extent, the speed of acquiring the recommendation result by the caller can be greatly increased, so that the use experience of the caller is improved.

The recommendation result determining module 30 is configured to determine a target recommendation result according to the recommendation request and the target recommendation module, and return the target recommendation result to the caller.

In the embodiment of the present invention, the recommendation result determining module 30 of the recommendation platform may process the recommendation request by using the determined target recommendation module, so as to obtain a corresponding target recommendation result, and return the target recommendation result to the corresponding caller. In a specific embodiment, the recommendation result determining module 30 may obtain the inference protocol interfaces corresponding to each target recommendation module, and sequentially call each inference protocol interface according to the expected recommendation flow, so as to call the corresponding target recommendation module to execute the corresponding module service functions such as recall or sequencing filtration, and further obtain the target recommendation result corresponding to the recommendation request, and then feed back the target recommendation result to the caller.

The embodiment of the invention provides a recommendation platform, which comprises the following components: the request acquisition module is used for acquiring a recommendation request of a calling party; the recommendation module determining module is used for determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist; and the recommendation result determining module is used for determining a target recommendation result according to the recommendation request and the target recommendation module and returning the target recommendation result to the calling party. According to the embodiment of the invention, when the recommendation platform does not have the corresponding target recommendation result of the recommendation request of the calling party, the corresponding target recommendation module is determined according to the recommendation request, and the final target recommendation result is determined by the recommendation request and the target recommendation module, so that the feedback efficiency of the recommendation result is improved to a certain extent, the personalized recommendation process supporting each recommendation request is realized, the recommendation result is more accurate, and the use experience is further effectively improved.

Example two

Fig. 2 is a schematic diagram of a recommendation platform according to a second embodiment of the present invention, which is further optimized and expanded based on the foregoing embodiments, and may be combined with each of the optional technical solutions in the foregoing embodiments. As shown in fig. 2, a recommendation platform provided in the second embodiment is further refined by the request acquisition module 10, the recommendation module determining module 20, and the recommendation result determining module 30, and the following specifically describes the structural composition of the recommendation platform in the second embodiment.

The request acquisition module 10 includes: a request receiving unit 11 and a request parsing unit 12.

A request receiving unit 11, configured to receive a recommendation request sent by a caller through a preset request manner; the preset request mode at least comprises the following steps: hypertext transfer protocol mode.

The preset request mode may refer to a network protocol used by the recommendation platform to provide services externally, and the preset request mode may at least include: hypertext transfer protocol (Hypertext Transfer Protocol, HTTP) mode.

In the embodiment of the present invention, the request receiving unit 11 may obtain the recommendation request sent by the caller through a preset request manner such as HTTP. It will be appreciated that in practical applications, other network protocols may be used as the preset request mode, for example, remote procedure call (Remote Procedure Call, RPC) may also be used, which is not limited in particular by the embodiment of the present invention.

A request parsing unit 12, configured to parse the recommendation request to obtain a corresponding target recommendation parameter; wherein the target recommendation parameters include at least one of: recommendation basis identification, expected recommendation flow, number of recommended results, list of excluded items, and filter condition identification.

In the embodiment of the present invention, the request parsing unit 12 may parse the obtained recommendation request to obtain the target recommendation parameter included in the recommendation request, where the target recommendation parameter may include at least one of the following: a recommendation basis identifier, an expected recommendation flow, a recommendation result number, an excluded item list and a filtering condition identifier, wherein the recommendation basis identifier is used for representing an identifier of a user or an item serving as a recommendation basis, such as an ID number, a name and the like of the user/item; the expected recommended flow is used for representing a character string of the expected recommended flow; the number of recommended results is used to specify the number of recommended results to be expected; the excluded item list is used to represent a list containing a series of excluded item identifiers; the filter term identifies an identifier for representing a predetermined filter term.

The recommendation module determining module 20 includes: a cache inquiry unit 21, a first target recommendation result determination unit 22, and a recommendation module determination unit 23.

The cache query unit 21 is configured to take a query identifier corresponding to the recommendation request as a query key, and detect whether the query key exists in the cache.

The query identifier may refer to a unique identifier that is encoded by the feature information of the recommendation request. The cache may be an embedded cache configured by the recommendation platform, which is a technology for storing data in a memory, and may provide effects of fast reading operation and reducing database load. A query key may refer to a unique key for storing and retrieving data blocks.

In the embodiment of the present invention, the cache query unit 21 may first use the query identifier corresponding to the recommendation request as the query key, and search whether the same query key exists in the cache configured by the recommendation platform, so as to determine whether the recommendation process needs to be executed in the following step.

The first target recommendation result determining unit 22 is configured to directly take recommendation data corresponding to the query key as a target recommendation result corresponding to the recommendation request if the query key exists.

In the embodiment of the present invention, if the first target recommendation result determining unit 22 detects that the corresponding query key exists in the cache, the recommendation data corresponding to the query key in the cache is used as the target recommendation result, and is directly returned to the caller, and the subsequent recommendation flow is skipped.

A recommendation module determining unit 23, configured to determine a target recommendation module according to an expected recommendation flow in the recommendation parameters corresponding to the recommendation request if the recommendation request does not exist; wherein the goal recommendation module includes at least one of: the target recall module and the target sorting module.

In the embodiment of the present invention, if the recommendation module determining unit 23 detects that there is no corresponding query key in the cache, that is, the cache misses, the recommendation module determining unit determines a corresponding target recommendation module according to an expected recommendation flow in target recommendation parameters obtained by resolving the recommendation request, where the target recommendation module includes at least one of the following: the target recall module is used for acquiring a recalled candidate list, and the target sorting module is used for sorting/filtering the candidate list so as to obtain a final target recommendation result.

The recommendation result determining module 30 includes: an inference interface acquisition unit 31, a first recommendation result generation unit 32, a second recommendation result generation unit 33, a second target recommendation result determination unit 34, and a recommendation result feedback unit 35.

The inference interface obtaining unit 31 is configured to obtain an inference protocol interface corresponding to each target recommendation module.

The inference protocol interface may be an interface configured by the recommendation platform in advance for each recommendation module to process an inference task, and the recommendation platform may call a corresponding recommendation module (or an instance of the recommendation module) by calling the inference protocol interface to execute a corresponding module service function.

In the embodiment of the present invention, the inference interface obtaining unit 31 may search, according to the information such as the module ID and the module name of each target recommendation module, the inference protocol interface corresponding to each target recommendation module in the inference protocol interfaces configured by the recommendation platform. In practical application, a configuration table may be configured, where the association relationship between the module ID of each recommendation module and the corresponding protocol interface is stored in the configuration table, so that when the recommendation process task is executed, the inference interface obtaining unit 31 may quickly obtain the corresponding inference protocol interface in the configuration table according to the module ID of each target recommendation module.

The first recommendation result generating unit 32 is configured to call each inference protocol interface according to the expected recommendation flow in the recommendation parameters corresponding to the target recommendation request, and generate a first recommendation result.

In the embodiment of the present invention, the first recommendation result generating unit 32 may call each inference protocol interface in turn according to the expected recommendation flow, so as to call the corresponding target recommendation module to execute the corresponding module service functions such as recall and/or sort, and further obtain the first recommendation result that is not filtered by the filtering condition.

The second recommendation result generating unit 33 is configured to determine a target preset filtering condition according to the filtering condition identifier in the target recommendation parameter corresponding to the recommendation request, and filter the first recommendation result into a second recommendation result by using the target preset filtering condition.

In the embodiment of the present invention, the second recommendation result generating unit 33 may obtain a corresponding target preset filtering condition in a position such as a filtering condition configuration table or a configuration file according to the filtering condition identifier in the target recommendation parameter obtained by analyzing the recommendation request, and then perform filtering processing on the obtained first recommendation result by using the target preset filtering condition to generate a corresponding second recommendation result.

And a second target recommendation result determining unit 34, configured to intercept or randomly complement the second recommendation result according to the number of recommendation results in the target recommendation parameters corresponding to the recommendation request, to obtain a target recommendation result.

In the embodiment of the present invention, the second target recommendation result determining unit 34 may intercept or randomly complement the second recommendation result according to the number of recommendation results in the target recommendation parameters obtained by analyzing the recommendation request, that is, intercept TOP-N in the second target recommendation result as the final target recommendation result if the number of the second target recommendation result exceeds the designated number of recommendation results N; if the number of the second target recommended results is less than the number N of the recommended results, randomly generating the remaining number of recommended results, and taking the second target recommended results and the randomly complemented recommended results together as the final target recommended results.

The recommendation result feedback unit 35 is configured to send the target recommendation result to the caller in a preset request manner.

In the embodiment of the present invention, after the target recommendation result is generated, the recommendation result feedback unit 35 may return the target recommendation result to the corresponding caller through a preset request manner such as HTTP, so as to complete the recommendation flow. Further, after the recommendation process is finished, the generated target recommendation result can be stored in the cache.

Further, on the basis of the embodiment of the present invention, the recommendation platform may further include: the random recommender is used for returning a preset number of random recommendation results when the target recommendation module is abnormal.

In the embodiment of the invention, the recommendation platform can be further configured with a random recommender, and when the target recommendation module in the recommendation process is abnormal or the recommendation module (or instance) is in an unavailable state due to training, the recommender can take over the corresponding target recommendation module so as to randomly generate a corresponding number of recommendation results.

The random recommender can provide the most basic availability for the recommendation platform, and when any process in the process of executing the recommendation flow task by the recommendation platform is abnormal, the recommendation platform almost can ensure that the available recommendation result is still returned, so that the use experience of a calling party is improved.

Further, on the basis of the embodiment of the present invention, the recommendation platform may further include: the data synchronization processing module is used for acquiring data change records of other service platforms connected with the recommendation platform, encoding and processing service change data corresponding to the data change records into integer records and storing the integer records into the target database table.

In the embodiment of the invention, the recommendation platform can be further configured with a data synchronization processing module, the data synchronization processing module can acquire the data change records of other service platforms connected with the recommendation platform, and after the data change records are acquired, the data synchronization processing module can encode and process the service change data corresponding to the data change records into integer records and store the integer records into the target database table, so that the integer records are directly used for training and reasoning tasks without additional preprocessing in the processes, and the training and reasoning speed of each recommendation module is greatly improved.

Further, on the basis of the embodiment of the present invention, the recommendation platform may further include: the data tracking module is used for acquiring tracking data in the platform business process in a preset data tracking mode; the preset data tracking mode at least comprises the following steps: the method comprises the steps of rotation log, recommended flow context model, application layer process tracking and returned data analysis, network and database request statistics and process performance tracking.

In the embodiment of the invention, in order to realize observability of the platform data, the recommendation platform can be further configured with a data tracking module, and the data tracking module can acquire the tracking data in the platform business process through a plurality of different preset data tracking modes, wherein the preset data tracking modes can at least comprise the following modes:

① Rotation log: the rotation log component configured by the data tracking module can select to divide log files and compress old logs according to fixed size or time interval according to deployment conditions and service flow conditions, and detail data in the platform service flow is recorded by using the classification multi-level rotation log;

② Recommending a flow context model: a context list containing a hash set can be created and maintained for each recommended flow, and the context list is used for recording the starting time, the execution state, the abnormal information and other data of each recommended flow;

③ Application layer process tracking and return data analysis: an event tracking component with a buffer area can be created and used for recording the starting time and time consumption of each application layer process and key information such as abnormal information, error information, recommendation result and the like of each step of the recommendation flow, and meanwhile, the information can be returned to a calling party for performing abnormal multi-angle accurate positioning;

④ Network and database request statistics: a statistics component with a configurable slot position can be created, and the statistics callback provided by the statistics component is used in the process of requesting each internal/external network and database, and the acquired time consumption and response condition are recorded in the corresponding slot position of the statistics component;

⑤ Process performance tracking: process performance tracking data may be obtained in real time via network requests.

Further, on the basis of the embodiment of the present invention, the recommendation platform may further include: the protocol interface maintenance module is used for maintaining protocol interfaces of all recommendation modules included in the recommendation platform; wherein, the protocol interface includes at least: a registration protocol interface, a training protocol interface, and an inference protocol interface.

In the embodiment of the present invention, the recommendation platform may be further configured with a protocol interface maintenance module, where the protocol interface maintenance module may be configured to perform maintenance on various protocol interfaces of all recommendation modules included in the recommendation platform, where the protocol interface may at least include: a registration protocol interface, a training protocol interface, and an inference protocol interface. In practical application, after a new algorithm/module is developed, key registration information of the new algorithm/module can be reported to the recommendation platform through a unified registration interface, for example, which interfaces of which link stages are realized, a monitoring address port, a training data preset content template type or a custom data column/data range, an updating frequency and the like. The training protocol interface may be used to invoke when the recommendation module needs to be trained.

Further, on the basis of the embodiment of the present invention, the recommendation platform may further include: the training updating module is used for creating a training updating task corresponding to the target instance of the recommending module according to the preset updating frequency and reporting an execution result corresponding to the training updating task.

In the embodiment of the invention, in order to ensure that each recommendation module of the recommendation platform can provide more accurate recommendation results, the recommendation platform can be further configured with a training update module for regularly carrying out training update on the examples of each recommendation module and reporting the execution results corresponding to the training update tasks to the recommendation platform, and if the training update tasks are successful, the corresponding recommendation model examples immediately enter an available state; if the training update task fails, the training update module records the corresponding error information and determines whether to restart the training update task.

The embodiment of the invention provides a recommendation platform, which is characterized in that after a recommendation request of a calling party is acquired, a cache configured by the recommendation platform is queried, and if a target recommendation result corresponding to the recommendation request exists in the cache, the target recommendation result in the cache is directly returned to the calling party, so that the speed of acquiring the recommendation result by the calling party is greatly increased; if the cache is not hit, the corresponding target recommendation module is determined according to the expected recommendation flow in the corresponding target recommendation parameters obtained by analyzing the recommendation request, and the final target recommendation result is determined by the recommendation request and the target recommendation module, so that the personalized recommendation flow supporting each recommendation request is realized, the recommendation result meets the requirements of the calling party, meanwhile, the effective recommendation result can be returned in almost any abnormal scene, and the use experience is effectively improved.

Example III

Fig. 3 is a schematic diagram of a recommendation platform according to a third embodiment of the present invention. As shown in fig. 3, the present embodiment combines techniques such as machine learning, software engineering, and micro-service development, and implements a recommendation platform supporting multiple stages and multiple flows and integrating data training and reasoning. The recommendation platform supports new algorithm/module/data content category through registration mode, and through the final data form required by the extensive investigation of the main stream machine learning algorithm, the preprocessing and characteristic engineering process is unified, and the new data is processed and stored in real time into the form required by the algorithm or the intermediate form easy to convert, thereby greatly improving the use experience of the online recommendation service and reducing the difficulty of system updating iteration.

As shown in fig. 4 and fig. 5, the present embodiment provides a recommendation platform using movie and television film sheet recommendation as a recommended service scene, where the recommendation platform uses HTTP to provide services to the outside and provides the most extensive applicability with the most popular network protocol; the typical communication protocol adopted inside the platform is RPC (remote procedure call), and realizes that a special data organization form and a data compression mode realized on a protobuf type system are contracted as extensions to improve the large-scale data transmission efficiency, and alternative protocols comprise rSocket, HTTP and the like. The recommendation platform is characterized in that an internal component is decoupled by a lightweight message queue middleware, and a protocol calling mode is adopted in a key performance process of the online service.

A recommendation platform as an online service will become an important part of the efficiency and quality of service of the business it supports. Thus, dynamic updating, exception handling, simplified computational flow, traceable status, and the ability to quickly respond to business demand changes are the most important features that the recommendation platform should have. The following specifically describes a recommendation platform provided in this embodiment:

(1) Algorithm/module registration

The embodiment defines a series of succinct and universal training (data updating)/reasoning protocol interfaces and a unified registration protocol interface for front and back link stage division of the current main stream of the recommendation platform.

When a new algorithm/module is developed, key information of the recommendation platform (such as which interfaces of which link stages are realized, which interfaces are monitored for address ports, preset content template types of training data or custom data columns/data ranges, update frequency and the like) is reported to the recommendation platform through a registration interface, the recommendation platform can store the acquired report information in a serialized mode, maintain proper connection according to the information and immediately try to perform data update operation on the report information, and then automatically establish training/data update tasks according to the designated update frequency.

When the recommendation platform receives a recommendation request containing the algorithm/module in a specified recommendation flow, whether the recommendation platform reports that the relevant protocol interface is realized or not can be checked according to the stored information, and if the check is successful, the connection is obtained or rebuilt and a proper function call is initiated by using a protocol stub.

(2) Online recommendation process

The recommendation flow is divided into two main phases, the front link that gets the candidate list and the back link that sorts/filters the candidate list.

A typical recommendation request contains an identifier of a user or an item on which the recommendation is based, a string that expresses the intended recommendation procedure, an integer that specifies the number of intended results, a list containing a series of excluded item identifiers, and a series of integers that represent predetermined filter term identifiers.

For the recommendation basis identifier in the recommendation request, the recommendation platform has a special accessor component to acquire the target value information from the correct process (such as cache, database or mathematical operation, etc.).

For the number of recommended results in the recommendation request, the platform sets the expected number of recommended results of each module according to the number of stages of the expected recommendation flow, the module ratio of each stage and the like.

For the expected recommendation flow in the recommendation request, the caller can flexibly specify different recommendation flows according to different service requirements, and the recommendation flows are exemplified as follows:

①Aa

②Aa+Ba

③[Aa,10],[Ab,45],[Ac,45]

④[Aa,30],[Ab,70]+Ba+Bb

Wherein A and B respectively represent a front link and a rear link, the front link is a multi-path parallel recall, the rear link is a module cascade to execute a sorting operation, [ Aa,10] represents that a module a of the front link outputs 10% of the number of expected recommended results, namely, 10 represents that the recommended results outputted by the front link module are in proportion to the number of expected recommended results.

The recommendation platform will make a pre-check to determine that invocation of the recommendation module is possible and necessary. For example when the following occurs: the number of the articles meeting the filtering condition is almost equal to or smaller than the expected recommended number, so that the starting of the recommendation flow is unnecessary, and the recommendation platform returns all recommendation results immediately.

The recommendation platform can correctly analyze the format agreed by each external interface and find the information of all relevant recommendation modules, such as an inference protocol interface and the like.

Before starting the recommendation flow, if the embedded cache of the application layer is configured to be started, the form characteristic information of the recommendation request is encoded into a unique identifier, the unique identifier is firstly used as a query key to query whether the cache has related target recommendation results, if the related target recommendation results are already present, the recommendation flow is skipped, and otherwise, the corresponding target recommendation results are stored in the cache after the recommendation flow is ended. Although the memory occupation of the whole recommendation platform can be improved to a certain extent, the speed of acquiring the recommendation result by the calling party can be greatly increased, and the use experience of the calling party is further improved.

After all condition checks pass and required information is prepared, a recommendation process is started, a front link and a rear link are constructed into a MapReduce process, a corresponding number of candidate lists (maps) are recalled simultaneously by each recommendation module of the front link, and after duplication removal, the recommendation modules are staggered and combined (Reduce) according to a certain rule, and the specific process is as follows: 1) If all the recommendation modules provide recommendation degree scoring information scaled to a preset range, ranking according to the scores; 2) If some recommendation modules do not provide recommendation degree scoring information, the candidate list provided by the recommendation module and the existing candidate list are crossed and placed into the final candidate list (such as [ a, b, c ] + [ x, y, z ] = [ a, x, b, y, c, z ]).

The above steps will make the current recommendation result satisfy the semantics of the intended recommendation flow in the recommendation request, which is not necessary if the intended recommendation flow contains a back link.

Each recommendation module of the back link is executed one by one in sequence, and the information required by each recommendation module is queried at a proper position before each recommendation module is executed; the output result of the last recommendation module in the expected recommendation flow is taken as the final target recommendation result of the recommendation flow, and the recommendation flow is ended.

If a certain recommending module of the front link in the recommending process is abnormal, the recommending result of the recommending module is replaced by a random result generated by a random recommender, and the abnormal is recorded in a recommending context flow recorder and a log system; if a certain recommendation module of the back link in the recommendation process is abnormal, the recommendation result of the previous stage is transmitted to the next stage as the recommendation result.

In addition, a sliding window based fusing algorithm is widely used throughout the recommendation platform. When an abnormality or overtime problem occurs continuously in a certain process in a short time, the recommendation platform can sense the whole phenomenon and cause the related process to fail rapidly; the processing mode after the quick failure is the same as the processing of the exception, the processing mode can be recorded in a corresponding log, and the problem that the related position in the recommendation platform is more serious can be effectively avoided.

The results of the actual recommended flow or the results taken from the cache will be filtered according to the filtering conditions expected by the request and then truncated or randomly complemented to the number specified by the request.

Meanwhile, the recommendation platform also supports the following special recommendation scenes: the caller inputs a recommended flow for the list of sheets to be ordered/filtered, in which case the front link will be skipped, for recommended scenarios requiring manual configuration or requiring special candidate item types.

(3) Random recommender, feature engineering ad hoc engagement and embedded caching

The random recommender can provide the most basic availability for the recommendation platform, and when any process in the process of executing the recommendation flow task by the recommendation platform is abnormal, the recommendation platform almost can ensure that the available recommendation result is still returned.

The random recommender is based on a special convention of platform data encoding: the serial numbers of all users/items/features are integers that are incremented from 1 to the total number, and this agreed-upon serial number and the global identifier of the user/item can be mapped bi-directionally by the accessor module; based on this convention, the random recommender can quickly generate recommendations by randomly intercepting and scrambling an incremental array of numbers from 1 to the total number of items.

The advantages of using the above convention also include: ① Almost all recommendation algorithms/modules receive integer items and user identifications, and quite many algorithms use successive indexes to query their pre-computed recommendation results or intermediate results from the order table/data; ② The serial number can be conveniently maintained by utilizing the self-increment id of the MySQL database and the like; ③ Smaller numbers facilitate compressed transmissions than random identifiers.

It will be appreciated that the hardware of the actual deployment scenario is generally poor, and some large algorithms still have a difficult-to-optimize computation time (e.g. more than 100 ms) in this scenario, and the recommendation platform provided in this embodiment uses a specially optimized embedded cache to store the preprocessing and final results, so that the total response time of the recommendation request can be reduced as much as possible compared to the common Redis-like remote dictionary service.

The buffer memory has configurable and unstable expiration time, and the 'configurable' improves the flexibility of the buffer memory along with the change of service scenes; an "unstable expiration time" can prevent to some extent a sudden increase in the number of actual flows (avalanches). The cache in the recommendation platform adopts a design of separating indexes from data, an index area is a hash set without a pointer, and a data area adopts a expandable circular queue as a storage structure; in golang languages used by the recommendation platform, the set without pointers can avoid the influence caused by garbage collection scanning, so that the related process has higher throughput rate and response speed; the circular queue is implemented based on a byte array that will likely provide compatibility with any structure of data due to the memory layout of the class c language.

The new entry will automatically contain a timestamp for expired eviction, an entry length for supporting dynamic data structures and fetching complete data, a non-hashed mapped key for detecting conflicts, and a header index for returning the entry for lookup.

(4) Online data processing and content registration

Recommending that the platform data be maintained in a sub-table in a MySQL database (or other database with similar functions); as an independent recommendation platform which is not dependent on or by other service platforms, the system at least comprises the following data synchronization modes:

① The platform database is used as a slave database to synchronize data tables related to other service platforms containing user or article information, and uses local script monitoring and near real-time processing of the synchronized table changes;

② The independent service directly serves as an update log for simulating remote pulling of related data tables from a database or obtains data from a data interface, and then processes data change;

③ Receiving data change records delivered by other service platforms by using Kafka or other message queue middleware with similar functions;

④ Data changes for the relevant service platforms are automatically monitored and synchronized across all service platforms using Databus or other similarly functioning data capture middleware.

After the recommendation platform acquires the data change records, the business change data corresponding to the data change records are respectively encoded and processed into integer records and stored in the corresponding tables of the platform database, so that the data are directly used for training and pushing, and additional preprocessing is hardly needed in the processes, thereby greatly improving the training and reasoning speed of each module/algorithm; if the platform is configured and started at the same time, the platform monitors the related data change records through the canary or other database synchronization middleware with similar functions, and updates related entries in the platform cache at proper time.

When the algorithm/module needs new content categories (e.g. feature columns, filtering conditions) as training or reasoning basis, due to: a) synchronous work is completed by external independent service or middleware, b) the cache supports dynamic data format, c) all data is integer data after preprocessing, and the recommendation platform can establish and store information required by inquiring the content through a declarative sql constructor only by calling descriptors (such as database name, table name and column name set) required by reporting and positioning the content category through a network interface.

When the algorithm/module registers to specify that this new feature class is required or when filtering according to this new condition class is specified in the request, the recommendation platform can query and cache such content data according to the previously constructed information and perform the transmission, filtering, etc. operations accordingly.

(5) Training/data update procedure

The data stored in the platform database table is already data suitable for the input model, the recommendation platform in the training process splices the data in batches according to the data requirement reported by each module/algorithm (recommendation module) in the registration, and the large-scale integer data after the batch of splicing is compressed varint and then transmitted to the corresponding module/algorithm, the module/algorithm can start the training/data updating task almost without any preprocessing, and the execution result is reported to the platform after the operation is completed.

During the training/data update process, the instance of the algorithm/module will enter an unavailable state to mask new recommendation requests; if the algorithm/module has other registered instances, these instances may still be used, otherwise the relevant recommendation request will be taken over by the random recommender.

If the recommendation platform receives a report that this process was successful, the algorithm/module instance will be immediately available; if this report is not received after a period of time, or an error report is received, the recommendation platform will record the error and decide whether to restart the process.

(6) Full flow observability

The recommendation platform realizes the full-flow observability in the platform business flow by adopting the following data tracking mode:

① Classified multi-level rotation log

The log file segmentation and old log compression can be selectively carried out according to fixed size or time interval by the log rotation component configured by the recommendation platform according to deployment conditions and service flow conditions, and the log file segmentation and old log compression are basic components of all components related to log files. In addition, each module of the recommendation platform comprises an independent global instance, and a developer can manually record detailed information in the process and store the detailed information in different files according to the category of the items; the log entry categories include, but are not limited to, debug, information, errors, statistics, and the like.

② Recommended flow context model

A corresponding context list containing a hash set may be created and maintained for each recommended flow for recording the start time, execution status, and exception information for each individual process in the recommended flow and formatting it for the caller and other tracking components.

③ Application layer process tracking and return data analysis

The recommendation platform uses an event tracking component with a buffer zone to record the starting time and time consumption of each application layer process and key information such as abnormal information, error information, recommendation result and the like of each step of the recommendation flow; and generating and recording trace id for each application process, wherein the id can correspond the tracking log of the process to the process record of the access log, the process context model and other positions, and return the tracking log to a calling party for carrying out abnormal multi-angle accurate positioning.

④ Network and database request statistics

The recommendation platform uses a statistics component with a configurable slot position, uses statistics callbacks provided by the statistics component in the process of requesting each internal/external network and database, and records time consumption and response conditions in the corresponding slot position of the component; a sliding window is used for carrying out stepwise automatic analysis and outputting the analysis to a log file; and simultaneously, a network interface is provided, so that the current state can be pulled from the statistics component in real time.

⑤ Process performance tracking data available in real time through network request

The recommendation platform may obtain process performance tracking data by: 1) allocs: sampling all memory allocations in the past; 2) Block: stack trace that causes congestion on synchronization primitives; 3) cmdline: a command line call of the current program; 4) goroutine: stack tracking for all current goroutine; 5) And (2) heap: sampling memory allocation of the movable object; 6) mutex: stack tracking by the contending mutex (lock) holder; 7) profile: CPU performance analysis, which can specify the duration of the tracking in parameters to acquire a tracking data file; 8) THREADCREATE: causing a stack trace to be created for the new operating system thread; 9) trace: the tracking of the current program execution may specify the duration of this tracking in the request parameters and obtain the tracking data file.

The recommendation platform provided by the third embodiment of the invention has the following beneficial effects:

① Any algorithm/module supporting any function is updated and online in real time without affecting the existing service;

② Automatically updating (training) the deep learning model;

③ The real-time specific data preprocessing mode and the characteristic engineering can quickly train reasoning and generate random results;

④ Decoupling from other product services;

⑤ Supporting a recommendation flow configurable for each request;

⑥ Full flow observability;

⑦ And the effective recommendation result is returned in almost any abnormal scene, so that the time is striven for troubleshooting while the use of related services is not influenced.

Example IV

Fig. 6 is a flowchart of a recommendation method according to a fourth embodiment of the present invention. As shown in fig. 6, the recommendation method provided in the fourth embodiment is applied to a recommendation platform, and specifically includes the following steps:

S410, acquiring a recommendation request of the calling party.

S420, when a target recommendation result corresponding to the recommendation request does not exist, determining a corresponding target recommendation module according to the recommendation request.

S430, determining a target recommendation result according to the recommendation request and the target recommendation module, and returning the target recommendation result to the calling party.

The specific implementation process of the recommendation method provided in this embodiment may refer to the above embodiment of the present invention, and will not be described herein.

The embodiment of the invention provides a recommendation method, which comprises the steps of obtaining a recommendation request of a calling party, determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist, determining a target recommendation result according to the recommendation request and the target recommendation module, and returning the target recommendation result to the calling party. According to the embodiment of the invention, when the recommendation platform does not have the corresponding target recommendation result of the recommendation request of the calling party, the corresponding target recommendation module is determined according to the recommendation request, and the final target recommendation result is determined by the recommendation request and the target recommendation module, so that the feedback efficiency of the recommendation result is improved to a certain extent, the personalized recommendation process supporting each recommendation request is realized, the recommendation result is more accurate, and the use experience is further effectively improved.

Example five

Fig. 7 shows a schematic diagram of an electronic device 50 that may be used to implement an embodiment of the invention. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic equipment may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.

As shown in fig. 7, the electronic device 50 includes at least one processor 51, and a memory such as a Read Only Memory (ROM) 52, a Random Access Memory (RAM) 53, etc. communicatively connected to the at least one processor 51, wherein the memory stores a computer program executable by the at least one processor, and the processor 51 may perform various suitable actions and processes according to the computer program stored in the Read Only Memory (ROM) 52 or the computer program loaded from the storage unit 58 into the Random Access Memory (RAM) 53. In the RAM 53, various programs and data required for the operation of the electronic device 50 can also be stored. The processor 51, the ROM 52 and the RAM 53 are connected to each other via a bus 54. An input/output (I/O) interface 55 is also connected to bus 54.

Various components in the electronic device 50 are connected to the I/O interface 55, including: an input unit 56 such as a keyboard, a mouse, etc.; an output unit 57 such as various types of displays, speakers, and the like; a storage unit 58 such as a magnetic disk, an optical disk, or the like; and a communication unit 59 such as a network card, modem, wireless communication transceiver, etc. The communication unit 59 allows the electronic device 50 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunications networks.

The processor 51 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 51 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, digital Signal Processors (DSPs), and any suitable processor, controller, microcontroller, etc. The processor 51 performs the various methods and processes described above, such as the recommended method.

In some embodiments, the recommendation method may be implemented as a computer program tangibly embodied on a computer-readable storage medium, such as the storage unit 58. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 50 via the ROM 52 and/or the communication unit 59. When the computer program is loaded into RAM 53 and executed by processor 51, one or more steps of the recommended methods described above may be performed. Alternatively, in other embodiments, the processor 51 may be configured to perform the recommendation method in any other suitable way (e.g. by means of firmware).

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

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

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

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

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

The computing system may include clients and servers. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service are overcome.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution of the present invention are achieved, and the present invention is not limited herein.

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

Claims (10)

1. A recommendation platform, the recommendation platform comprising:

the request acquisition module is used for acquiring a recommendation request of a calling party;

The recommendation module determining module is used for determining a corresponding target recommendation module according to the recommendation request when a target recommendation result corresponding to the recommendation request does not exist;

And the recommendation result determining module is used for determining the target recommendation result according to the recommendation request and the target recommendation module and returning the target recommendation result to the calling party.

2. The recommendation platform of claim 1, wherein the request acquisition module comprises:

the request receiving unit is used for receiving the recommendation request sent by the calling party in a preset request mode; the preset request mode at least comprises the following steps: a hypertext transfer protocol mode;

the request analysis unit is used for analyzing the recommendation request to obtain corresponding target recommendation parameters; wherein the target recommendation parameter comprises at least one of: recommendation basis identification, expected recommendation flow, number of recommended results, list of excluded items, and filter condition identification.

3. The recommendation platform of claim 1, wherein the recommendation module determination module comprises:

The cache query unit is used for taking a query identifier corresponding to the recommendation request as a query key and detecting whether the query key exists in a cache;

The first target recommendation result determining unit is used for directly taking recommendation data corresponding to the query key as a target recommendation result corresponding to the recommendation request if the query key exists;

A recommendation module determining unit, configured to determine, if the recommendation request does not exist, the target recommendation module according to an expected recommendation flow in target recommendation parameters corresponding to the recommendation request; wherein the target recommendation module comprises at least one of the following: the target recall module and the target sorting module.

4. The recommendation platform according to claim 1, wherein said recommendation result determining module comprises:

the reasoning interface acquisition unit is used for acquiring the reasoning protocol interface corresponding to each target recommendation module;

the first recommendation result generating unit is used for calling each reasoning protocol interface according to the expected recommendation flow in the recommendation parameters of the target corresponding to the recommendation request to generate a first recommendation result;

The second recommendation result generation unit is used for determining a target preset filtering condition according to the filtering condition identification in the target recommendation parameters corresponding to the recommendation request, and filtering the first recommendation result into a second recommendation result by utilizing the target preset filtering condition;

The second target recommendation result determining unit is used for intercepting or randomly complementing the second recommendation result into the target recommendation result according to the recommendation result number in the target recommendation parameter corresponding to the recommendation request;

and the recommendation result feedback unit is used for sending the target recommendation result to the calling party in a preset request mode.

5. The recommendation platform of claim 1, further comprising: the random recommendation device is used for returning a preset number of random recommendation results when the target recommendation module is abnormal.

6. The recommendation platform of claim 1, further comprising: and the data synchronization processing module is used for acquiring data change records of other service platforms connected with the recommendation platform, encoding and processing service change data corresponding to the data change records into integer records and storing the integer records into a target database table.

7. The recommendation platform of claim 1, further comprising: the data tracking module is used for acquiring tracking data in the platform business process in a preset data tracking mode; the preset data tracking mode at least comprises the following steps: log rotation, recommended flow context model, application layer process tracking and return data analysis, network and database request statistics and process performance tracking.

8. A recommendation method, applied to a recommendation platform, comprising:

Acquiring a recommendation request of a calling party;

when a target recommendation result corresponding to the recommendation request does not exist, determining a corresponding target recommendation module according to the recommendation request;

And determining the target recommendation result according to the recommendation request and the target recommendation module, and returning the target recommendation result to the calling party.

9. An electronic device, the electronic device comprising:

At least one processor; and

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

The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the recommendation method of claim 8.

10. A computer readable storage medium storing computer instructions for causing a processor to execute the recommendation method of claim 8.