Object recommendation method and device in cold start scene
By obtaining and matching semantic ID sequences in cold start scenarios, the problem of low accuracy and efficiency of cold start object recommendations is solved, and more accurate and efficient object recommendations are achieved.
Patent Information
- Application Number
- CN202411365529.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-02-07
AI Technical Summary
In cold start scenarios, objects in cold start state cannot achieve accurate recommendations of objects due to lack of interactive data or fewer interactive data.
By obtaining the initial semantic ID sequence of the initial recommendation object, obtaining the target semantic ID sequence according to the frequency of occurrence of the semantic ID sequence, and matching the target semantic ID sequence with the candidate semantic ID sequence of the candidate recommendation object, obtaining the corresponding candidate recommendation object as the target Recommended person.
It improves the accuracy and efficiency of object recommendation in cold start scenarios, can better understand the deep-level characteristics of the initial recommendation object and candidate recommendation object, and improves the acquisition speed and generalization of the target recommendation object.
Smart Images
Figure CN119398059A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of Internet technology, and in particular to the field of artificial intelligence technology such as large language models, deep learning, cloud services, big data, etc. A method, device, electronic device, and readable storage medium for object recommendation in a cold start scenario are provided. Background Art
[0002] Traditional object recommendation methods generally recommend corresponding objects to users based on the interaction data between users and objects. However, this recommendation method relies on the interaction data between users and objects. In the cold start scenario, objects in the cold start state lack interaction data or have little interaction data, so accurate object recommendations cannot be achieved. Summary of the invention
[0003] According to the first aspect of the present disclosure, a method for object recommendation in a cold start scenario is provided, comprising: obtaining an initial semantic ID sequence based on an initial recommended object, wherein the initial recommended object is a recommended object in a non-cold start state; obtaining a target semantic ID sequence from the initial semantic ID sequence based on the frequency of occurrence of the semantic ID sequence; obtaining a candidate semantic ID sequence based on candidate recommended objects, wherein the candidate recommended objects are recommended objects in a cold start state; matching the target semantic ID sequence with the candidate semantic ID sequence, obtaining a candidate recommended object corresponding to the matched candidate semantic ID sequence as a target recommended object; and making recommendations based on the target recommended object.
[0004] According to a second aspect of the present disclosure, an object recommendation device in a cold start scenario is provided, comprising: a first acquisition unit, used to acquire an initial semantic ID sequence based on an initial recommended object, wherein the initial recommended object is a recommended object in a non-cold start state; a processing unit, used to acquire a target semantic ID sequence from the initial semantic ID sequence according to the frequency of occurrence of the semantic ID sequence; a second acquisition unit, used to acquire a candidate semantic ID sequence based on a candidate recommended object, wherein the candidate recommended object is a recommended object in a cold start state; a matching unit, used to match the target semantic ID sequence with the candidate semantic ID sequence, and acquire the candidate recommended object corresponding to the matched candidate semantic ID sequence as the target recommended object; and a recommendation unit, used to make recommendations based on the target recommended object.
[0005] According to a third aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.
[0006] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method as described above.
[0007] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the method as described above when executed by a processor.
[0008] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.
[0010] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;
[0011] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;
[0012] Figure 3 is a schematic diagram according to a third embodiment of the present disclosure;
[0013] Figure 4 The block diagram is a block diagram of an electronic device used to implement the object recommendation method in the cold start scenario of the embodiment of the present disclosure. DETAILED DESCRIPTION
[0014] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and mechanisms is omitted in the following description.
[0015] Figure 1 Schematic diagram of the first embodiment of the present disclosure. Figure 1 As shown, the object recommendation method in the cold start scenario of this embodiment specifically includes the following steps:
[0016] S101, acquiring an initial semantic ID sequence according to an initial recommended object, wherein the initial recommended object is a recommended object in a non-cold start state;
[0017] S102, acquiring a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence;
[0018] S103, obtaining a candidate semantic ID sequence according to a candidate recommendation object, wherein the candidate recommendation object is a recommendation object in a cold start state;
[0019] S104, matching the target semantic ID sequence with the candidate semantic ID sequence, and obtaining a candidate recommended object corresponding to the matched candidate semantic ID sequence as a target recommended object;
[0020] S105: Make recommendations based on the target recommendation object.
[0021] The object recommendation method in the cold start scenario of this embodiment obtains the target recommended object from the candidate recommended objects for recommendation based on the initial semantic ID sequence corresponding to the initial recommended object and the candidate semantic ID sequence corresponding to the candidate recommended objects. Since the semantic ID sequence introduced in the object recommendation process of this embodiment can characterize the deep-level characteristics of the recommended object, the recommendation system can better understand the initial recommended object and the candidate recommended object based on the semantic ID sequence, thereby improving the accuracy of the target recommended object determined based on the initial recommended object, and further improving the accuracy of object recommendation in the cold start scenario.
[0022] In addition, since the semantic ID sequence is composed of discrete semantic identifiers (i.e., semantic IDs), this embodiment can also improve the acquisition speed of the target recommended object when obtaining the target recommended object corresponding to the initial recommended object through the discrete semantic ID sequence, thereby improving the object recommendation efficiency in the cold start scenario.
[0023] In this embodiment, the semantic ID sequence (Semantic ID Sequence) is a string of semantic identifiers (semantic IDs) obtained by transforming a high-dimensional semantic feature vector through a preset quantization or encoding method. The discrete semantic IDs (i.e., semantic identifiers) in the semantic ID sequence are used to represent the processed semantic information; the deep-level features represented by the semantic ID sequence include not only the surface attribute information of the recommended object, but also the deeper semantic meaning of the recommended object.
[0024] In this embodiment, the recommended objects include but are not limited to commodity objects, audio objects, video objects, text information objects, etc.
[0025] In this embodiment, the recommendation object in the cold start state is a newly added recommendation object in the recommendation system, and the newly added recommendation object lacks interaction data with the user; the recommendation object in the non-cold start state is a recommendation object that has existed in the recommendation system for a period of time; the object recommendation method in the cold start scenario of this embodiment is used to recommend the recommendation objects in the cold start state.
[0026] When executing S101, this embodiment may first obtain an initial recommended object, and then obtain an initial semantic ID sequence based on the obtained initial recommended object; the number of initial recommended objects obtained when executing S101 in this embodiment may be one or more.
[0027] When executing S101 to obtain the initial recommended object, this embodiment can obtain at least one preset recommended object in a non-cold start state, such as multiple popular recommended objects in a non-cold start state, as at least one initial recommended object; or it can obtain only one recommended object in a non-cold start state corresponding to the user characteristics of the current user (such as user behavior characteristics, user attribute characteristics, user preference characteristics, etc.) as at least one initial recommended object.
[0028] In this embodiment, when executing S101 to obtain the initial semantic ID sequence corresponding to the initial recommended object, the semantic ID sequence corresponding to the initial recommended object obtained in advance in the offline stage may be obtained as the initial semantic ID sequence.
[0029] If the semantic ID sequence of the initial recommended object is not obtained in advance in the offline stage, this embodiment may include the following contents when executing S101 to obtain the initial semantic ID sequence according to the initial recommended object: obtaining object description information corresponding to the initial recommended object. The object description information in this embodiment may be the title of the recommended object, the label of the recommended object, etc.; obtaining the semantic feature vector corresponding to the initial recommended object based on the obtained object description information. This embodiment may input the object description information into a pre-trained large language model, and obtain the semantic feature vector corresponding to the initial recommended object based on the output result of the large language model; obtaining the initial semantic ID sequence based on the semantic feature vector corresponding to the initial recommended object.
[0030] Among them, when executing S101, this embodiment can use a preset quantization (for example, multi-layer residual quantization) or encoding method to quantize or encode the semantic feature vector corresponding to the initial recommended object to obtain an initial semantic ID sequence; or the semantic feature vector of the initial recommended object can be input into a pre-trained sequence conversion model (for example, a pre-trained residual quantization variational autoencoder model) to obtain the initial semantic ID sequence according to the output result of the sequence conversion model.
[0031] The initial semantic ID sequence obtained by executing S101 in this embodiment has a preset length, and the size of the semantic IDs at different positions in the sequence is between 1 and 256; for example, if the preset length is 4, the initial semantic ID sequence of the initial recommended object 1 can be [1, 12, 22, 40], and different semantic IDs are used to reflect different semantic meanings; the preset length in this embodiment can be any positive integer greater than or equal to 3.
[0032] It is understandable that if this embodiment executes S101 to obtain multiple initial recommended objects, the initial semantic ID sequence corresponding to each initial recommended object can be obtained respectively, that is, the number of the obtained initial semantic ID sequences is the same as the number of initial recommended objects.
[0033] After executing S101 to obtain the initial semantic ID sequence, this embodiment executes S102 to obtain a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence; the number of target semantic ID sequences obtained in this embodiment can be one or more.
[0034] Specifically, when executing S102 of this embodiment to obtain the target semantic ID sequence from the initial semantic ID sequence according to the frequency of occurrence of the semantic ID sequence, the implementation method that can be adopted is: sorting the initial semantic ID sequence in order of occurrence frequency from high to low, for example, sorting multiple initial semantic ID sequences; obtaining the initial semantic ID sequence ranked in the top N positions as the target semantic ID sequence, where N is a positive integer greater than or equal to 1.
[0035] That is to say, this embodiment screens at least one initial semantic ID sequence according to the frequency of occurrence, filters out initial semantic ID sequences with less frequency of occurrence, and selects initial semantic ID sequences with more frequency of occurrence as at least one target semantic ID sequence, thereby improving the accuracy of the acquired target semantic ID sequence.
[0036] For example, if at least one initial semantic ID sequence includes sequence 1 [1, 12, 22], sequence 2 [3, 20, 33], sequence 3 [1, 12, 22], sequence 4 [5, 12, 30], sequence 5 [1, 12, 22] and sequence 6 [5, 12, 30], then the occurrence frequency of sequence [1, 12, 22] is 3, the occurrence frequency of sequence [3, 20, 33] is 1, and the occurrence frequency of sequence [5, 12, 30] is 2; if N is 2, the target semantic ID sequences obtained by executing S102 in this embodiment are [1, 12, 22] and [5, 12, 30].
[0037] In actual application scenarios, in order to ensure that the semantic ID sequence can more accurately reflect the semantic information of the recommended object, the preset length of the initial semantic ID sequence is usually relatively long, for example, the preset length is 6; since the initial semantic ID sequence includes relatively many semantic IDs, when directly obtaining the target semantic ID sequence based on the initial semantic ID sequence, there will be problems such as slow acquisition speed and low generalization degree.
[0038] Therefore, when executing S102 to obtain the target semantic ID sequence from the initial semantic ID sequence according to the frequency of occurrence of the semantic ID sequence, this embodiment can also adopt the following implementation method: according to the preset sequence position, obtain the initial semantic ID subsequence corresponding to the initial semantic ID sequence, and the length of the initial semantic ID subsequence in this embodiment is less than the length of the initial semantic ID sequence; sort the initial semantic ID subsequences in order of occurrence frequency from high to low; obtain the initial semantic ID subsequences ranked in the top M positions as the target semantic ID sequence, where M is a positive integer greater than or equal to 1; in this embodiment, M and N may be the same or different.
[0039] That is to say, this embodiment screens at least one initial semantic ID subsequence according to the frequency of occurrence of the initial semantic ID subsequence corresponding to the initial semantic ID sequence, and uses the initial semantic ID subsequence with a higher frequency of occurrence as at least one target semantic ID sequence. Since the sequence length of the initial semantic ID subsequence is smaller than the sequence length of the initial semantic ID sequence, the amount of data that needs to be processed during sorting is reduced. Therefore, this embodiment can improve the speed of acquiring the target semantic ID sequence, and by acquiring the target semantic ID sequence based on the initial semantic ID subsequence with a smaller sequence length, it can also improve the generalization degree of the acquired at least one target semantic ID sequence.
[0040] Among them, when executing S102 to obtain the initial semantic ID subsequence corresponding to the initial semantic ID sequence according to the preset sequence position, this embodiment can first obtain the semantic ID corresponding to the preset sequence position from the initial semantic ID sequence for the current initial semantic ID sequence, and then obtain the initial semantic ID subsequence corresponding to the initial semantic ID based on the obtained semantic ID.
[0041] This embodiment does not limit the number of preset sequence positions. For example, if the number of preset sequence positions is two, the two preset sequence positions may be adjacent or non-adjacent.
[0042] For example, if the initial semantic ID sequence is [1, 12, 3, 30], if the preset sequence positions are the first position and the second position, the initial semantic ID subsequence obtained by executing S102 in this embodiment is [1, 12]; if the preset sequence positions are the first position and the third position, the initial semantic ID subsequence obtained by executing S102 in this embodiment is [1, 3].
[0043] After executing S102 to obtain the target semantic ID sequence, this embodiment executes S103 to obtain a candidate semantic ID sequence according to the candidate recommendation objects; in this embodiment, the candidate recommendation objects are recommendation objects in a cold start state.
[0044] When executing S103, this embodiment may first obtain candidate recommendation objects, and then obtain candidate semantic ID sequences based on the obtained candidate recommendation objects; the number of candidate recommendation objects obtained when executing S103 in this embodiment may be one or more.
[0045] The candidate semantic ID sequence obtained by executing S103 in this embodiment has a preset length, and the size of the semantic IDs at different positions in the sequence is between 1 and 256; wherein the candidate semantic ID sequence has the same sequence length as the initial semantic ID sequence.
[0046] When executing S103, this embodiment may obtain at least one recommended object in a cold start state from a preset index library as at least one candidate recommended object.
[0047] The preset index library of this embodiment, in addition to including at least one recommended object in a cold start state, also includes a semantic ID sequence corresponding to each recommended object, that is, this embodiment can obtain at least one candidate semantic ID sequence based on the semantic ID sequence obtained in advance in the offline stage.
[0048] When executing S103, this embodiment may obtain a semantic ID sequence corresponding to at least one candidate recommendation object from a preset index library as at least one candidate semantic ID sequence.
[0049] If the candidate semantic ID sequence cannot be obtained from the preset index library, when executing S103 to obtain the candidate semantic ID sequence according to the candidate recommendation object, the implementation method that can be adopted in this embodiment is: obtain the object description information corresponding to the candidate recommendation object. The object description information in this embodiment can be the title of the recommended object, the label of the recommended object, etc.; obtain the semantic feature vector corresponding to the candidate recommendation object based on the obtained object description information. This embodiment can input the object description information into a pre-trained large language model, and obtain the semantic feature vector of the candidate recommendation object based on the output result of the large language model; and obtain the candidate semantic ID sequence based on the semantic feature vector corresponding to the candidate recommendation object.
[0050] It is understandable that if this embodiment executes S103 to obtain multiple candidate recommendation objects, the candidate semantic ID sequence corresponding to each candidate recommendation object can be obtained respectively, that is, the number of obtained candidate semantic ID sequences is the same as the number of candidate recommendation objects.
[0051] After executing S103 to obtain the candidate semantic ID sequence, this embodiment executes S104 to match the target semantic ID sequence with the candidate semantic ID sequence, and obtains the candidate recommended object corresponding to the matched candidate semantic ID sequence as the target recommended object.
[0052] When executing S104 in this embodiment, if there are multiple target semantic ID sequences or multiple candidate semantic ID sequences, each target semantic ID sequence is matched with each candidate semantic ID sequence respectively, and the candidate semantic ID sequence determined to match the current target semantic ID sequence is the same candidate semantic ID sequence as the current target semantic ID sequence.
[0053] When executing S102 , this embodiment further obtains a target semantic ID sequence according to the initial semantic ID subsequence obtained at the preset sequence position, that is, the sequence length of the target semantic ID sequence is different from the sequence length of the candidate semantic ID sequence.
[0054] In order to improve the matching accuracy between semantic ID sequences, this embodiment may also adopt the following implementation method when executing S104 to match the target semantic ID sequence with the candidate semantic ID sequence: when it is determined that the sequence length of the target semantic ID sequence is different from the sequence length of the candidate semantic ID sequence, obtain the candidate semantic ID subsequence corresponding to the candidate semantic ID sequence according to the preset sequence position; match the target semantic ID sequence with the candidate semantic ID subsequence, and obtain the candidate semantic ID sequence corresponding to the matched candidate semantic ID subsequence as the candidate semantic ID sequence that matches the target semantic ID sequence.
[0055] That is to say, this embodiment can match the target semantic ID sequence with the candidate semantic ID subsequence obtained according to at least one preset sequence position when the target semantic ID sequence and the candidate semantic ID sequence have different sequence lengths, thereby ensuring that the matched sequences have the same sequence length, thereby improving the matching accuracy between the semantic ID sequences.
[0056] It can be understood that if this embodiment executes S104 to obtain multiple candidate semantic ID subsequences matching the target semantic ID sequence, multiple target recommendation objects can be obtained according to the multiple candidate semantic ID sequences.
[0057] In this embodiment, after executing S104 to obtain the target recommendation object, S105 is executed to make a recommendation based on the obtained target recommendation object.
[0058] When executing S105 , this embodiment may recommend the acquired at least one target recommendation object to the current user in a random recommendation order.
[0059] In order to further improve the recommendation accuracy of the target recommended object, when executing S105 to make recommendations based on the acquired target recommended object, this embodiment can also adopt the following implementation method: obtain the sorting result of the target recommended object according to the candidate semantic ID sequence corresponding to the target recommended object; and recommend the target recommended objects in sequence according to the sorting result.
[0060] That is to say, this embodiment introduces the semantic ID sequence corresponding to the recommended objects to sort the target recommended objects, which can improve the richness of the information used in the sorting, thereby improving the accuracy of the obtained sorting results and the recommendation efficiency of the target recommended objects.
[0061] Among them, when this embodiment executes S105 to obtain the ranking result of the target recommendation object according to the candidate semantic ID sequence corresponding to the target recommendation object, the candidate semantic ID sequences corresponding to different target recommendation objects can be input into the pre-trained ranking model respectively, and then the ranking result of the target recommendation object can be obtained according to the output result of the ranking model.
[0062] After executing S105 to obtain the sorting results, this embodiment can recommend different target recommendation objects to the current user in turn according to the order of each target recommendation object in the sorting results. For example, when the current user is watching a video in the form of a feed stream, as the current user refreshes the feed, different target recommendation objects (i.e., video objects) will be displayed to the current user in turn.
[0063] Figure 2 is a schematic diagram according to the second embodiment of the present disclosure. Figure 2 As shown, when executing S105 "obtaining a ranking result of the target recommended object according to the candidate semantic ID sequence corresponding to the target recommended object", this embodiment may also include the following contents:
[0064] S201, obtaining user characteristics of the current user and object characteristics corresponding to the target recommended object;
[0065] S202. Obtain a ranking result of the target recommended object according to the user features, the object features corresponding to the target recommended object, and the candidate semantic ID sequence corresponding to the target recommended object.
[0066] That is to say, when sorting the target recommended objects, this embodiment, in addition to using the candidate semantic ID sequence corresponding to the target recommended objects, also uses the user features corresponding to the current user and the object features corresponding to the target recommended objects, enriching the information used in the sorting as much as possible, thereby improving the accuracy of the sorting results of at least one target recommended object obtained.
[0067] In this embodiment, the user characteristics of the current user may include the user behavior characteristics, user attribute characteristics, user preference characteristics, etc. of the current user; the object characteristics corresponding to the target recommended object may include object title characteristics, object tag characteristics, object author characteristics, etc.
[0068] When executing S202, this embodiment can input user features, object features corresponding to the target recommended object and the candidate semantic ID sequence into a pre-trained sorting model, thereby obtaining the sorting result of the target recommended object according to the output result of the sorting model.
[0069] When executing S202 to obtain the ranking result of the target recommended object according to the user characteristics, the object characteristics corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object, the present embodiment may also adopt the following implementation method: obtaining the object fusion characteristics corresponding to the target recommended object according to the object characteristics corresponding to the target recommended object and the candidate semantic ID sequence; obtaining the ranking result of the target recommended object according to the user characteristics, the object characteristics corresponding to the target recommended object, the candidate semantic ID sequence corresponding to the target recommended object and the object fusion characteristics corresponding to the target recommended object.
[0070] That is to say, this embodiment can also fuse the object features corresponding to the target recommended object with the candidate semantic ID sequence, and then combine the obtained object fusion features to sort at least one target recommended object, which can further improve the accuracy of the obtained sorting results.
[0071] Among them, when executing S202, this embodiment can input user characteristics, object characteristics corresponding to the target recommended object, candidate semantic ID sequence corresponding to the target recommended object, and object fusion characteristics corresponding to the target recommended object into a pre-trained sorting model, and obtain the sorting result of the target recommended object according to the output result of the sorting model.
[0072] Figure 3 is a schematic diagram according to the third embodiment of the present disclosure. Figure 3 As shown, the object recommendation device 300 in the cold start scenario of this embodiment includes:
[0073] A first acquisition unit 301 is used to acquire an initial semantic ID sequence according to an initial recommended object, where the initial recommended object is a recommended object in a non-cold start state;
[0074] The processing unit 302 is used to obtain a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence;
[0075] The second acquisition unit 303 is used to acquire a candidate semantic ID sequence according to a candidate recommendation object, wherein the candidate recommendation object is a recommendation object in a cold start state;
[0076] A matching unit 304 is used to match the target semantic ID sequence with the candidate semantic ID sequence, and obtain a candidate recommended object corresponding to the matched candidate semantic ID sequence as a target recommended object;
[0077] The recommendation unit 305 is used to make recommendations based on the target recommendation object.
[0078] In this embodiment, the recommended objects include but are not limited to commodity objects, audio objects, video objects, text information objects, etc.
[0079] The first acquisition unit 301 may first acquire at least one initial recommended object, and then acquire at least one initial semantic ID sequence according to the at least one initial recommended object.
[0080] When acquiring at least one initial recommended object, the first acquisition unit 301 can acquire a preset plurality of recommended objects in a non-cold start state as at least one initial recommended object; it can also acquire a plurality of recommended objects in a non-cold start state corresponding to user characteristics of the current user (such as user behavior characteristics, user attribute characteristics, user preference characteristics, etc.) as at least one initial recommended object.
[0081] When acquiring at least one initial semantic ID sequence, the first acquiring unit 301 may acquire a semantic ID sequence corresponding to at least one initial recommended object obtained in advance in an offline stage as the at least one initial semantic ID sequence.
[0082] If the semantic ID sequence of the initial recommended object is not obtained in advance in the offline stage, the first acquisition unit 301 may include the following contents when acquiring the initial semantic ID sequence based on the initial recommended object: acquiring object description information corresponding to the initial recommended object; acquiring a semantic feature vector corresponding to the initial recommended object based on the acquired object description information; and acquiring the initial semantic ID sequence based on the semantic feature vector corresponding to the initial recommended object.
[0083] Among them, the first acquisition unit 301 can use a preset quantization (for example, multi-layer residual quantization) or encoding method to quantize or encode the semantic feature vector corresponding to the initial recommended object to obtain an initial semantic ID sequence; or the semantic feature vector of the initial recommended object can be input into a pre-trained sequence conversion model (for example, a pre-trained residual quantization variational autoencoder model) to obtain the initial semantic ID sequence according to the output result of the sequence conversion model.
[0084] The initial semantic ID sequence acquired by the first acquisition unit 301 has a preset length, and the sizes of the semantic IDs at different positions in the sequence are between 1 and 256.
[0085] In this embodiment, after the first acquisition unit 301 acquires the initial semantic ID sequence, the processing unit 302 acquires the target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence.
[0086] Specifically, when the processing unit 302 obtains the target semantic ID sequence from the initial semantic ID sequence according to the frequency of occurrence of the semantic ID sequence, the implementation method that can be adopted is: sorting the initial semantic ID sequence in order from high to low frequency of occurrence; obtaining the initial semantic ID sequence ranked in the top N positions as the target semantic ID sequence, where N is a positive integer greater than or equal to 1.
[0087] That is to say, this embodiment screens at least one initial semantic ID sequence according to the frequency of occurrence, filters out initial semantic ID sequences with less frequency of occurrence, and selects initial semantic ID sequences with more frequency of occurrence as target semantic ID sequences, thereby improving the accuracy of the acquired target semantic ID sequence.
[0088] In actual application scenarios, in order to ensure that the semantic ID sequence can more accurately reflect the semantic information of the recommended object, the preset length of the initial semantic ID sequence is usually relatively long, for example, the preset length is 6; since the initial semantic ID sequence includes relatively many semantic IDs, when directly obtaining the target semantic ID sequence based on the initial semantic ID sequence, there will be problems such as slow acquisition speed and low generalization degree.
[0089] When the processing unit 302 obtains the target semantic ID sequence from the initial semantic ID sequence according to the frequency of occurrence of the semantic ID sequence, the following implementation method can also be adopted: according to the preset sequence position, an initial semantic ID subsequence corresponding to the initial semantic ID sequence is obtained, and the length of the initial semantic ID subsequence in this embodiment is less than the length of the initial semantic ID sequence; the initial semantic ID subsequences are sorted in order from high to low according to the frequency of occurrence; and the initial semantic ID subsequences ranked in the top M positions are obtained as the target semantic ID sequence, where M is a positive integer greater than or equal to 1.
[0090] That is to say, this embodiment screens at least one initial semantic ID subsequence according to the frequency of occurrence of the initial semantic ID subsequence corresponding to the initial semantic ID sequence, and takes the initial semantic ID subsequence with a higher frequency of occurrence as the target semantic ID sequence. Since the sequence length of the initial semantic ID subsequence is smaller than the sequence length of the initial semantic ID sequence, the amount of data that needs to be processed during sorting is reduced. Therefore, this embodiment can improve the speed of acquiring the target semantic ID sequence, and by acquiring the target semantic ID sequence based on the initial semantic ID subsequence with a smaller sequence length, it can also improve the generalization degree of at least one target semantic ID sequence acquired.
[0091] Among them, when the processing unit 302 obtains the initial semantic ID subsequence corresponding to the initial semantic ID sequence according to the preset sequence position, for each initial semantic ID sequence, it can first obtain the semantic ID corresponding to the preset sequence position from the initial semantic ID sequence, and then obtain the initial semantic ID subsequence corresponding to the initial semantic ID based on the obtained semantic ID.
[0092] This embodiment does not limit the number of preset sequence positions. For example, if the number of preset sequence positions is two, the two preset sequence positions may be adjacent or non-adjacent.
[0093] In this embodiment, after the processing unit 302 obtains the target semantic ID sequence, the second obtaining unit 303 obtains the candidate semantic ID sequence according to the candidate recommendation objects; in this embodiment, the candidate recommendation objects are recommendation objects in a cold start state.
[0094] The second acquisition unit 303 may first acquire at least one candidate recommendation object, and then acquire at least one candidate semantic ID sequence according to the at least one candidate recommendation object.
[0095] The candidate semantic ID sequence acquired by the second acquisition unit 303 has a preset length, and the sizes of the semantic IDs at different positions in the sequence are between 1 and 256; wherein the candidate semantic ID sequence has the same sequence length as the initial semantic ID sequence.
[0096] The second acquisition unit 303 may acquire at least one recommended object in a cold start state from the preset index library as at least one candidate recommended object.
[0097] The preset index library of this embodiment, in addition to including at least one recommended object in a cold start state, also includes a semantic ID sequence corresponding to each recommended object, that is, this embodiment can obtain at least one candidate semantic ID sequence based on the semantic ID sequence obtained in advance in the offline stage.
[0098] The second acquisition unit 303 may acquire, from a preset index library, a semantic ID sequence corresponding to at least one candidate recommendation object, as at least one candidate semantic ID sequence.
[0099] If the candidate semantic ID sequence cannot be obtained from the preset index library, the second acquisition unit 303 may obtain the candidate semantic ID sequence based on the candidate recommended object by: obtaining object description information corresponding to the candidate recommended object; obtaining the semantic feature vector corresponding to the candidate recommended object based on the obtained object description information; and obtaining the candidate semantic ID sequence based on the semantic feature vector corresponding to the candidate recommended object.
[0100] In this embodiment, after the second acquisition unit 303 acquires the candidate semantic ID sequence, the matching unit 304 matches the target semantic ID sequence with the candidate semantic ID sequence, and acquires the candidate recommended object corresponding to the matched candidate semantic ID sequence as the target recommended object.
[0101] The candidate semantic ID sequence determined by the matching unit 304 to match the current target semantic ID sequence is the same candidate semantic ID sequence as the current target semantic ID sequence.
[0102] In order to improve the matching accuracy between semantic ID sequences, the matching unit 304 can also adopt the following implementation method when matching the target semantic ID sequence with the candidate semantic ID sequence: when it is determined that the sequence length of the target semantic ID sequence is different from the sequence length of the candidate semantic ID sequence, obtain the candidate semantic ID subsequence corresponding to the candidate semantic ID sequence according to the preset sequence position; match the target semantic ID sequence with the candidate semantic ID subsequence, and obtain the candidate semantic ID sequence corresponding to the matched candidate semantic ID subsequence as the candidate semantic ID sequence that matches the target semantic ID sequence.
[0103] That is to say, this embodiment can match the target semantic ID sequence with the candidate semantic ID subsequence obtained according to at least one preset sequence position when the target semantic ID sequence and the candidate semantic ID sequence have different sequence lengths, thereby ensuring that the matched sequences have the same sequence length, thereby improving the matching accuracy between the semantic ID sequences.
[0104] In this embodiment, after the matching unit 304 obtains the target recommendation object, the recommendation unit 305 makes a recommendation based on the obtained target recommendation object.
[0105] The recommendation unit 305 may recommend the acquired at least one target recommendation object to the current user in a random recommendation order.
[0106] In order to further improve the recommendation accuracy of the target recommended object, the recommendation unit 305 can also adopt the following implementation method when making recommendations based on the acquired target recommended object: obtain the sorting result of the target recommended object according to the candidate semantic ID sequence corresponding to the target recommended object; and recommend the target recommended objects in sequence according to the sorting result.
[0107] That is to say, this embodiment introduces the semantic ID sequence corresponding to the recommended objects to sort the target recommended objects, which can improve the richness of the information used in the sorting, thereby improving the accuracy of the obtained sorting results and the recommendation efficiency of the target recommended objects.
[0108] Among them, when the recommendation unit 305 obtains the ranking result of the target recommendation object according to the candidate semantic ID sequence corresponding to the target recommendation object, the candidate semantic ID sequences corresponding to different target recommendation objects can be input into the pre-trained ranking model respectively, and then the ranking result of the target recommendation object can be obtained according to the output result of the ranking model.
[0109] After obtaining the sorting results, the recommendation unit 305 can recommend different target recommendation objects to the current user in turn according to the order of each target recommendation object in the sorting results. For example, when the current user is watching a video in the form of a feed stream, as the current user refreshes, different target recommendation objects (i.e., video objects) will be displayed to the current user in turn.
[0110] When the recommendation unit 305 obtains the ranking result of the target recommended object according to the candidate semantic ID sequence corresponding to the target recommended object, it may also include the following contents: obtaining the user characteristics of the current user and the object characteristics corresponding to the target recommended object; obtaining the ranking result of the target recommended object according to the user characteristics, the object characteristics corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object.
[0111] That is to say, when sorting the target recommended objects, this embodiment, in addition to using the candidate semantic ID sequence corresponding to the target recommended objects, also uses the user features corresponding to the current user and the object features corresponding to the target recommended objects, enriching the information used in the sorting as much as possible, thereby improving the accuracy of the sorting results of at least one target recommended object obtained.
[0112] In this embodiment, the user characteristics of the current user may include the user behavior characteristics, user attribute characteristics, user preference characteristics, etc. of the current user; the object characteristics corresponding to the target recommended object may include object title characteristics, object tag characteristics, object author characteristics, etc.
[0113] The recommendation unit 305 can input user features, object features corresponding to the target recommended object, and a candidate semantic ID sequence corresponding to the target recommendation into a pre-trained sorting model, thereby obtaining a sorting result of the target recommended object based on an output result of the sorting model.
[0114] When the recommendation unit 305 obtains the ranking result of the target recommended object based on the user characteristics, the object characteristics corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object, the following implementation method can also be adopted: obtaining the object fusion characteristics corresponding to the target recommended object based on the object characteristics corresponding to the target recommended object and the candidate semantic ID sequence; obtaining the ranking result of the target recommended object based on the user characteristics, the object characteristics corresponding to the target recommended object, the candidate semantic ID sequence corresponding to the target recommended object and the object fusion characteristics corresponding to the target recommended object.
[0115] That is to say, this embodiment can also fuse the object features corresponding to the target recommended object with the candidate semantic ID sequence, and then combine the obtained object fusion features to sort at least one target recommended object, which can further improve the accuracy of the obtained sorting results.
[0116] Among them, the recommendation unit 305 can input user features, object features corresponding to the target recommended object, the candidate semantic ID sequence corresponding to the target recommended object, and the object fusion features corresponding to the target recommended object into a pre-trained sorting model, and obtain the sorting result of the target recommended object according to the output result of the sorting model.
[0117] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0118] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.
[0119] like Figure 4 , is a block diagram of an electronic device according to an object recommendation method in a cold start scenario according to an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.
[0120] like Figure 4As shown, the device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the device 400 can also be stored. The computing unit 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.
[0121] A number of components in the device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the device 400 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0122] The computing unit 401 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 401 performs the various methods and processes described above, such as the object recommendation method in the cold start scenario. For example, in some embodiments, the object recommendation method in the cold start scenario can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 408.
[0123] In some embodiments, part or all of the computer program may be loaded and / or installed on the device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded into the RAM 403 and executed by the computing unit 401, one or more steps of the object recommendation method in the cold start scenario described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the object recommendation method in the cold start scenario in any other appropriate manner (e.g., by means of firmware).
[0124] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0125] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable cold-start scenario object recommendation device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.
[0126] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0127] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0128] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.
[0129] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server for a distributed system, or a server combined with a blockchain.
[0130] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.
[0131] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. An object recommendation method in a cold start scenario, comprising: According to the initial recommended object, an initial semantic ID sequence is obtained, wherein the initial recommended object is a recommended object in a non-cold start state; According to the occurrence frequency of the semantic ID sequence, a target semantic ID sequence is obtained from the initial semantic ID sequence; According to the candidate recommendation objects, a candidate semantic ID sequence is obtained, wherein the candidate recommendation objects are recommendation objects in a cold start state; Matching the target semantic ID sequence with the candidate semantic ID sequence, and obtaining a candidate recommended object corresponding to the matched candidate semantic ID sequence as a target recommended object; Recommendations are made according to the target recommendation object.
2. The method according to claim 1, wherein: The step of acquiring a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence includes: Sorting the initial semantic ID sequence in descending order of occurrence frequency; The first N initial semantic ID sequences are obtained as the target semantic ID sequence, where N is a positive integer greater than or equal to 1.
3. The method according to claim 1, wherein: The step of acquiring a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence includes: According to the preset sequence position, obtaining the initial semantic ID subsequence corresponding to the initial semantic ID sequence; Sorting the initial semantic ID subsequences in descending order of occurrence frequency; The initial semantic ID subsequences ranked in the first M positions are obtained as the target semantic ID sequence, where M is a positive integer greater than or equal to 1.
4. The method according to claim 1, wherein: The matching of the target semantic ID sequence with the candidate semantic ID sequence comprises: When it is determined that the sequence length of the target semantic ID sequence is different from the sequence length of the candidate semantic ID sequence, obtaining a candidate semantic ID subsequence corresponding to the candidate semantic ID sequence according to a preset sequence position; Matching the target semantic ID sequence with the candidate semantic ID subsequences respectively; A candidate semantic ID sequence corresponding to the matched candidate semantic ID subsequence is obtained as a candidate semantic ID sequence that matches the target semantic ID sequence.
5. The method according to claim 1, wherein: The making a recommendation according to the target recommendation object comprises: Obtaining a ranking result of the target recommended object according to a candidate semantic ID sequence corresponding to the target recommended object; According to the ranking result, the target recommendation object is recommended.
6. The method according to claim 5, wherein: The obtaining, according to the candidate semantic ID sequence corresponding to the target recommended object, a ranking result of the target recommended object comprises: Obtaining user characteristics of the current user and object characteristics corresponding to the target recommended object; A ranking result of the target recommended object is obtained according to the user features, the object features corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object.
7. The method according to claim 6, wherein: The obtaining the ranking result of the target recommended object according to the user feature, the object feature corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object comprises: According to the object features corresponding to the target recommended object and the candidate semantic ID sequence, obtaining the object fusion features corresponding to the target recommended object; A ranking result of the target recommended object is obtained according to the user features, the object features corresponding to the target recommended object, the candidate semantic ID sequence corresponding to the target recommended object and the object fusion features corresponding to the target recommended object.
8. The method according to claim 1, wherein: The obtaining of the initial semantic ID sequence according to the initial recommended object comprises: Obtaining object description information corresponding to the initially recommended object; According to the object description information, obtaining a semantic feature vector corresponding to the initial recommended object; The initial semantic ID sequence is obtained according to the semantic feature vector corresponding to the initial recommended object.
9. The method according to claim 1, wherein: The step of obtaining a candidate semantic ID sequence according to the candidate recommendation object includes: Obtaining object description information corresponding to the candidate recommendation object; According to the object description information, obtaining a semantic feature vector corresponding to the candidate recommendation object; The candidate semantic ID sequence is obtained according to the semantic feature vector corresponding to the candidate recommendation object.
10. An object recommendation device in a cold start scenario, comprising: A first acquisition unit is used to acquire an initial semantic ID sequence according to an initial recommended object, wherein the initial recommended object is a recommended object in a non-cold start state; A processing unit, configured to obtain a target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence; A second acquisition unit is used to acquire a candidate semantic ID sequence according to a candidate recommendation object, wherein the candidate recommendation object is a recommendation object in a cold start state; A matching unit, used for matching the target semantic ID sequence with the candidate semantic ID sequence, and obtaining a candidate recommended object corresponding to the matched candidate semantic ID sequence as a target recommended object; A recommendation unit is used to make recommendations based on the target recommendation object.
11. The device according to claim 10, wherein: When the processing unit obtains the target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence, the processing unit specifically performs: Sorting the initial semantic ID sequence in descending order of occurrence frequency; The first N initial semantic ID sequences are obtained as the target semantic ID sequence, where N is a positive integer greater than or equal to 1.
12. The device according to claim 10, wherein: When the processing unit obtains the target semantic ID sequence from the initial semantic ID sequence according to the occurrence frequency of the semantic ID sequence, the processing unit specifically performs: According to the preset sequence position, obtaining the initial semantic ID subsequence corresponding to the initial semantic ID sequence; Sorting the initial semantic ID subsequences in descending order of occurrence frequency; The initial semantic ID subsequences ranked in the first M positions are obtained as the target semantic ID sequence, where M is a positive integer greater than or equal to 1.
13. The device according to claim 10, wherein: When matching the target semantic ID sequence with the candidate semantic ID sequence, the matching unit specifically performs: When it is determined that the sequence length of the target semantic ID sequence is different from the sequence length of the candidate semantic ID sequence, obtaining a candidate semantic ID subsequence corresponding to the candidate semantic ID sequence according to a preset sequence position; Matching the target semantic ID sequence with the candidate semantic ID subsequence; A candidate semantic ID sequence corresponding to the matched candidate semantic ID subsequence is obtained as a candidate semantic ID sequence that matches the target semantic ID sequence.
14. The device according to claim 10, wherein: When the recommendation unit makes a recommendation based on the target recommendation object, the recommendation unit specifically performs: Obtaining a ranking result of the target recommended object according to a candidate semantic ID sequence corresponding to the target recommended object; According to the ranking result, the target recommendation object is recommended.
15. The device according to claim 14, wherein: When the recommendation unit obtains the ranking result of the target recommendation object according to the candidate semantic ID sequence corresponding to the target recommendation object, the recommendation unit specifically performs: Obtaining user characteristics of the current user and object characteristics of the target recommended object; A ranking result of the target recommended object is obtained according to the user features, the object features corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object.
16. The device according to claim 15, wherein: When the recommendation unit obtains the ranking result of the target recommended object according to the user feature, the object feature corresponding to the target recommended object and the candidate semantic ID sequence corresponding to the target recommended object, the recommendation unit specifically performs: According to the object features corresponding to the target recommended object and the candidate semantic ID sequence, obtaining the object fusion features corresponding to the target recommended object; A ranking result of the target recommended object is obtained according to the user features, the object features corresponding to the target recommended object, the candidate semantic ID sequence corresponding to the target recommended object and the object fusion features corresponding to the target recommended object.
17. The device according to claim 10, wherein: When the first acquisition unit acquires the initial semantic ID sequence according to the initial recommended object, the first acquisition unit specifically performs: Obtaining object description information corresponding to the initially recommended object; According to the object description information, obtaining a semantic feature vector corresponding to the initial recommended object; The initial semantic ID sequence is obtained according to the semantic feature vector corresponding to the initial recommended object.
18. The device according to claim 10, wherein: When the second acquisition unit acquires the candidate semantic ID sequence according to the candidate recommendation object, the second acquisition unit specifically performs: Obtaining object description information corresponding to the candidate recommendation object; According to the object description information, obtaining a semantic feature vector corresponding to the candidate recommendation object; The candidate semantic ID sequence is obtained according to the semantic feature vector corresponding to the candidate recommendation object.
19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.
20. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 9.
21. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.