Event recommendation method based on user interaction sequence and related device

Abstract

The application belongs to the field of electronic information technology, and discloses an event recommendation method based on user interaction sequences and related devices, which comprises the following steps: obtaining a user to be recommended and each event to be recommended, extracting the recommendation influence features of the user to be recommended, and extracting the feature representation vectors and user interaction sequences of each event to be recommended; obtaining the participation prediction probability of the user to be recommended for each event to be recommended by calling a preset deep interaction perception event recommendation model according to the recommendation influence features of the user to be recommended, the feature representation vectors and user interaction sequences of each event to be recommended; and recommending each event to be recommended to the user to be recommended in descending order of the participation prediction probability. By introducing the event user interaction sequence as a feature, the modeling capability of the model for event semantics is improved, the user interaction sequence reflects the past interests of the user, and the accuracy of event recommendation is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of electronic information technology and relates to an event recommendation method and related apparatus based on user interaction sequences. Background Technology

[0002] Event-based social networks provide internet users with a new way to socialize: posting social activity information online and participating in social events offline. With the rise of event-based social networks, the number of newly created and posted event plans on these platforms is rapidly increasing daily. Users are finding it increasingly difficult to find events that match their interests and preferences, leading to information overload. To help users quickly discover events that meet their personal preferences, event recommendation research has become a key focus in the field of event-based social networks. Existing event recommendation methods face two main problems: 1. Difficulty in accurately capturing the semantic changes in events brought about by new user participation. 2. The cold start problem for newly created and less popular events.

[0003] Existing event recommendation methods struggle to model the semantic changes in events brought about by new user participation, impacting the accuracy and effectiveness of recommendations. For instance, classic event recommendation methods based on matrix factorization and probabilistic matrix factorization calculate ratings using trained latent vector representations of users and events. However, these latent vectors do not change with the participating users, making it difficult to model semantic changes. Furthermore, content- and context-based event recommendation methods often fail to capture subtle shifts in features when faced with semantic changes caused by new user participation, as only a few relevant features change.

[0004] To address the cold start problem for emerging and niche events, a common approach is to characterize events using their side information instead of relying solely on unique identifiers. However, this method has two serious drawbacks: First, the event embedding is randomly initialized during cold start. Directly using this embedding for calculations and recommendations will significantly reduce the accuracy of the model's recommendations. Second, the event embedding only reaches its optimal state after acquiring a large number of interacting users, which is difficult for most events with a small number of participating users, thus leading to reduced recommendation performance. In real-world event-based social networks, most events have a relatively small number of participating users, which negatively impacts the overall effectiveness of event recommendations. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an event recommendation method and related apparatus based on user interaction sequences.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] In a first aspect, the present invention provides an event recommendation method based on user interaction sequences, comprising:

[0008] Obtain the users to be recommended and each event to be recommended, extract the recommendation impact features of the users to be recommended, and extract the feature representation vectors and user interaction sequences of each event to be recommended;

[0009] Based on the recommendation influence characteristics of the users to be recommended, as well as the feature representation vectors of each event to be recommended and the user interaction sequence, the predicted participation probability of the users to be recommended for each event to be recommended is obtained by calling the preset deep interaction perception event recommendation model.

[0010] Recommend each event to be recommended to the user in descending order of the probability of participating in the prediction.

[0011] Optionally, the extraction of feature representation vectors for each event to be recommended includes:

[0012] Obtain the recommendation impact features of each event to be recommended; encode the numerical recommendation impact features of each event to be recommended using the CLIP model to obtain the first representation vector of each event to be recommended; encode the categorical recommendation impact features of each event to be recommended using a pre-trained embedding vector matrix to obtain the second representation vector of each event to be recommended; concatenate the first and second representation vectors of each event to obtain the recommendation impact features of each event to be recommended.

[0013] Optionally, the recommendation influence characteristics of the users to be recommended include one or more of the following: age, gender, city, and tag preference; the recommendation influence characteristics of the events to be recommended include one or more of the following: participating age, participating gender, number of participating users, participating city, participating week, participating distance, and tag preference.

[0014] Optionally, when extracting the feature representation vector of each event to be recommended, if the event to be recommended is a new event or a niche event, the feature representation vector of the event to be recommended is obtained in the following way:

[0015] The feature representation vector of the historical interaction events of the group to which the event to be recommended belongs is averaged using the following formula to obtain the feature representation vector of the event to be recommended:

[0016]

[0017] in, Let E be the feature representation vector of the event to be recommended, Agg be the shift-invariant aggregation function, and E be the feature representation vector of the event to be recommended. gtx is a set of historical events belonging to the group to which the event to be recommended belongs. t For E gt The feature representation vector of event t.

[0018] Optionally, the deep interaction-aware event recommendation model includes an embedded representation layer, an attention-based user interaction sequence perception layer, and an output layer based on a multilayer perceptron.

[0019] The embedding representation layer encodes the numerical recommendation influence features of the user to be recommended using the CLIP model to obtain the first representation vector of the user to be recommended; it encodes the categorical recommendation influence features of the user to be recommended using a pre-trained embedding vector matrix to obtain the second representation vector of the user to be recommended; the first and second representation vectors of the user to be recommended are concatenated to obtain the feature representation vector of the user to be recommended; the user interaction sequence perception layer obtains the feature representation vector of each user in the user interaction sequence of each event to be recommended, and performs interaction operations between the feature representation vector of the user to be recommended and the feature representation vector of each user in the user interaction sequence of each event to be recommended to obtain the interaction vector between the user to be recommended and each event to be recommended; the output layer obtains the participation prediction probability of the user to be recommended for each event to be recommended based on the feature representation vector of each event to be recommended and the interaction vector between the user to be recommended and each event to be recommended.

[0020] Optionally, the attention-based user interaction sequence perception layer specifically comprises:

[0021]

[0022]

[0023] Where, x interact Let be the interaction vector between the user to be recommended and the event to be recommended, where Att represents the attention mechanism, and u is the feature representation vector of the user to be recommended. Let be the feature representation vector of user i in the user interaction sequence of the event to be recommended, m be the length of the user interaction sequence of the event to be recommended, a(*,*) represent the attention coefficient generated by the attention mechanism of two vectors of the same length, which is a scalar; Concate represents the vector concatenation operation, and MLP represents the multilayer perceptron.

[0024] The output layer based on the multilayer perceptron is specifically:

[0025] h 0 =Concate(x interact ,x1,x2,...,x n )

[0026] h 1=Relu(W 1 h 0 +b 1 )

[0027] h 2 =Relu(W 2 h 1 +b 2 ) ...

[0029]

[0030] Where, x n Let h be the nth feature representation vector, k be the number of layers in the multilayer perceptron, and sigmoid and ReLU are both non-linear activation functions. -1 Let be the state vector of the (k-1)th layer perceptron. W is the output vector. k Let b be the parameter of the k-th layer perceptron. k Let n be the bias vector of the k-th layer perceptron, and n be the number of feature representation vectors.

[0031] Optionally, the deep interaction-aware event recommendation model uses the following loss function during training:

[0032]

[0033] Where l is the loss, N is the number of samples, (x,y) is a single sample, D is the sample set, and y is the true value. The output vector is the predicted value.

[0034] A second aspect of the present invention provides an event recommendation system based on user interaction sequences, comprising:

[0035] The data acquisition module is used to acquire users to be recommended and each event to be recommended, extract the recommendation impact features of users to be recommended, and extract the feature representation vectors and user interaction sequences of each event to be recommended.

[0036] The model processing module is used to obtain the predicted probability of the user's participation in each recommended event by calling a preset deep interaction-aware event recommendation model, based on the recommendation influence characteristics of the user to be recommended, the feature representation vector of each recommended event, and the user interaction sequence.

[0037] The event recommendation module is used to recommend events to users in descending order of their probability of participation in the prediction.

[0038] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described event recommendation method based on user interaction sequences.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described event recommendation method based on user interaction sequences.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] This invention presents an event recommendation method based on user interaction sequences. First, it acquires the recommendation influence features of the user to be recommended, as well as the feature representation vectors and user interaction sequences of each event to be recommended. Then, it calls a pre-defined deep interaction-aware event recommendation model to obtain the predicted participation probability of the user for each event, thus transforming the recommendation problem into a participation probability problem. Events are then recommended to the user in descending order of predicted participation probability. By introducing the event user interaction sequence as a feature, which reflects the user's past interests, considering these interests improves the accuracy of current event recommendations and enhances the model's ability to model event semantics. It allows for deep interaction between the user to be recommended and all participating users of the event to address semantic changes brought about by new user participation, effectively capturing the semantic changes caused by new user involvement and ensuring the quality of event recommendations. Attached Figure Description

[0042] Figure 1 This is a flowchart of an event recommendation method based on user interaction sequences according to an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram illustrating the relationship between whether a user participates in an event and the number of mutual friends among the users who have participated in the event, according to an embodiment of the present invention.

[0044] Figure 3 This is a schematic diagram illustrating the relationship between whether a user participates in an event and the group to which the event belongs, according to an embodiment of the present invention.

[0045] Figure 4 This is a schematic diagram illustrating the average similarity between event embedding representations on a real dataset, according to an embodiment of the present invention.

[0046] Figure 5 This is a schematic diagram illustrating the number of overlapping users across events on different datasets in an embodiment of the present invention.

[0047] Figure 6This is a schematic diagram illustrating the principle of the deep interactive perception event recommendation model in an embodiment of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0049] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0050] The present invention will now be described in further detail with reference to the accompanying drawings:

[0051] See Figure 1 In one embodiment of the present invention, an event recommendation method based on user interaction sequences is provided, comprising the following steps:

[0052] S1: Obtain the users to be recommended and each event to be recommended, extract the recommendation impact features of the users to be recommended, and extract the feature representation vectors and user interaction sequences of each event to be recommended.

[0053] S2: Based on the recommendation influence characteristics of the users to be recommended, the feature representation vectors of each event to be recommended, and the user interaction sequence, the predicted participation probability of the users to be recommended for each event to be recommended is obtained by calling the preset deep interaction perception event recommendation model.

[0054] S3: Recommend each event to be recommended to the user in descending order of the probability of participating in the prediction.

[0055] This invention presents an event recommendation method based on user interaction sequences. First, it acquires the recommendation influence features of the user to be recommended, as well as the feature representation vectors and user interaction sequences of each event to be recommended. Then, it calls a pre-defined deep interaction-aware event recommendation model to obtain the predicted participation probability of the user for each event, thus transforming the recommendation problem into a participation probability problem. Events are then recommended to the user in descending order of predicted participation probability. By introducing the event user interaction sequence as a feature, which reflects the user's past interests, considering these interests improves the accuracy of current event recommendations and enhances the model's ability to model event semantics. It allows for deep interaction between the user to be recommended and all participating users of the event to address semantic changes brought about by new user participation, effectively capturing the semantic changes caused by new user involvement and ensuring the quality of event recommendations.

[0056] In one possible implementation, the step of extracting the feature representation vectors of each event to be recommended includes: obtaining the recommendation impact features of each event to be recommended; encoding the numerical recommendation impact features of each event to be recommended using a CLIP (Contrastive Language-Image Pre-training) model to obtain a first representation vector of each event to be recommended; encoding the categorical recommendation impact features of each event to be recommended using a pre-trained embedding vector matrix to obtain a second representation vector of each event to be recommended; and concatenating the first and second representation vectors of each event to be recommended to obtain the recommendation impact features of each event to be recommended.

[0057] In one possible implementation, the recommendation influence characteristics of the user to be recommended include one or more of the following: age, gender, city, and tag preference; the recommendation influence characteristics of the event to be recommended include one or more of the following: participating age, participating gender, number of participating users, participating city, participating week, participating distance, and tag preference.

[0058] Specifically, based on the working mode of event-based social network application platforms, this study analyzed the impact of social interaction on user participation in events, constructing features such as age, gender, and the number of participating users; it also analyzed the impact of time and space on user participation in events, constructing features such as city, weekday, and participation distance; and it further analyzed the impact of tag preferences on user participation in events, constructing features such as tag preferences and user interaction sequences.

[0059] See Figures 2 to 3This study explores factors influencing user engagement, such as social, temporal and spatial factors, and tag preferences, providing rich feature inputs for subsequent modules. Unlike recommendations in scenarios like products, news, and music, the mutual influence between users in event-based social networks has a more profound impact on user behavior. Figure 2 This shows the relationship between whether a user participates in an event and the number of mutual friends among those who have already participated. It reveals that participating users have an average of 1.3 friends on the event, while non-participating users have only about 0.3% of their friends. The relationship between a user and a group (a group of users with similar interests) typically refers to membership; users are more likely to participate in events organized by groups they have already joined. Figure 3 The analysis shows the relationship between user participation in events and the groups to which the events belong. It reveals that approximately 50% of the events selected by users are from their own groups, highlighting the importance of groups in event-based social networking applications. Other features are analyzed using the same method (see Table 1) to obtain the final usable features.

[0060] Table 1 Event Mining Feature Table

[0061] Feature type user event Context numerical values age gender Number of participating users category City Week Participation distance Multiple categories Tag preferences Tag preferences User interaction sequence

[0062] In one possible implementation, when extracting the feature representation vector of each event to be recommended, if the event to be recommended is a new event or a niche event, the feature representation vector of the event to be recommended is obtained as follows: the feature representation vectors of the historical interaction events of the group to which the event to be recommended belongs are averaged using the following formula to obtain the feature representation vector of the event to be recommended:

[0063]

[0064] in, Let E be the feature representation vector of the event to be recommended, Agg be the shift-invariant aggregation function, and E be the feature representation vector of the event to be recommended. gt x is a set of historical events belonging to the group to which the event to be recommended belongs. t For E gt The feature representation vector of event t.

[0065] Specifically, the cold start problem in event recommendation is addressed by leveraging the similarity between events within a group, i.e., recommending emerging and less popular events. A group is composed of users with similar interests within an event social network; therefore, events hosted by the same group are considered to have high similarity. To mitigate the cold start problem in event recommendation by utilizing the similarity of events within a group, a group dependency initialization strategy uses the embedding representations (feature vectors) of historical events hosted by the group to generate feature vectors for new events. This more accurate feature vector is then substituted into the model, improving recommendation accuracy.

[0066] In this embodiment, the above operation is referred to as the Group Attachment Initialization Strategy (GAIS). GAIS can alleviate the performance degradation caused by random initialization of the embedded representation of initial events. Its core idea is to utilize the similarity between events within a group. See also... Figures 4 to 5 , Figure 4 The average similarity between event embeddings on different datasets is shown, and it can be seen that the average similarity of events within the same group is about twice that of events in different groups. Figure 5 The data shows the number of overlapping users between events across different datasets. It can be seen that if two events belong to the same group, the average number of participating users is around 2.0, significantly higher than the number of events across different groups. This indicates that events within the same group also exhibit similarity among their participants. Groups are composed of users with similar interests and preferences within an event social network; therefore, events hosted by the same group have high similarity. The group dependency initialization strategy utilizes the embeddings of historical events hosted by the group to generate the embeddings of new events.

[0067] Traditional random initialization strategies embed all events into a single vector and initialize them as randomly generated vectors. This approach, to some extent, reduces the model's recommendation performance for cold-start events. This invention employs a group-dependent initialization strategy to alleviate the cold-start problem in recommending unpopular and emerging events. It uses average pooling of the embeddings of all events within a group as the embedding for new events. By incorporating a better embedding for new events into the model, the recommendation of new events becomes more accurate.

[0068] In one possible implementation, the deep interaction-aware event recommendation model includes an embedding representation layer, an attention-based user interaction sequence perception layer, and an output layer based on a multilayer perceptron. The embedding representation layer encodes numerical recommendation influence features from the recommendation influence features of the user to be recommended using a CLIP model to obtain a first representation vector of the user to be recommended; it then encodes categorical recommendation influence features from the recommendation influence features of the user to be recommended using a pre-trained embedding vector matrix to obtain a second representation vector of the user to be recommended; finally, it concatenates the first and second representation vectors of the user to be recommended to obtain a feature representation vector of the user to be recommended. The user interaction sequence perception layer obtains the feature representation vectors of each user in the user interaction sequence of each event to be recommended, and performs interaction operations between the feature representation vectors of the user to be recommended and the feature representation vectors of each user in the user interaction sequence of each event to be recommended to obtain an interaction vector between the user to be recommended and each event to be recommended. The output layer calculates the participation prediction probability of the user to be recommended for each event to be recommended based on the feature representation vectors of each event to be recommended and the interaction vectors between the user to be recommended and each event to be recommended.

[0069] See Figure 6 The event recommendation problem is transformed into a classification problem of whether users will participate in the event. The recommended events are ranked by predicting the event ratings, and the prediction process uses the following formula:

[0070]

[0071] Where f and θ are the functions from the sample to the model's predicted score and their corresponding parameters, respectively, y∈{0,1} represents whether the user participates in the event, and e i , and s e These represent the event identifier, feature set, and user interaction sequence, respectively. i and These represent the user's identifier and feature set, respectively. The most significant feature of the deep interaction-aware event recommendation model is that it retains the user interaction sequence as input. This allows subsequent modules to deeply interact with both the users who have already participated in the event and the target user. The model sensitively captures semantic changes brought about by new user participation through an attention mechanism, thereby improving the final recommendation performance.

[0072] The deep interaction-aware event recommendation model consists of three parts: 1) Embedded representation layer: different strategies are designed for different feature types to be input into the subsequent model; 2) User interaction sequence awareness layer: users in the event's user interaction sequence are deeply interacted with through an attention mechanism to obtain richer event semantic representations; 3) Output layer: the obtained embedded representations are interacted with and mapped into output probabilities.

[0073] Embedding representation layer: For numerical text features, the text features encoded by the CLIP model are directly used to obtain their representation vectors; for categorical features, the embedding representation layer generates a trainable embedding vector matrix and obtains the corresponding vectors as their feature representations.

[0074] User interaction sequence perception layer: This layer employs an attention-based mechanism to deeply interact with the user to be recommended and all participating users of the recommended event, addressing the semantic changes brought about by the user's participation. It calculates similarity scores between the sequence events and the current target event, and then uses these similarity scores and past sequence events to calculate the user's past interests.

[0075] The user's historical event sequence is used as input, allowing subsequent modules to deeply interact with both participating users and target users. The formula for the interaction awareness layer is as follows:

[0076]

[0077] Where, x interact Let be the interaction vector between the user to be recommended and the event to be recommended, where Att represents the attention mechanism, and u is the feature representation vector of the user to be recommended. Let be the feature representation vector of user i in the user interaction sequence of the event to be recommended, m be the length of the user interaction sequence of the event to be recommended, and a(*,*) represent the attention coefficient generated by the attention mechanism of two vectors of the same length, which is a scalar.

[0078] Under the attention mechanism, the target user representation directly interacts with all user representations in the event interaction sequence, resulting in a deep interaction and an output vector that serves as input to the subsequent model. This vector is highly sensitive to the users in the interaction sequence and possesses a strong ability to express event semantics. The output vector of this part also changes as the user interaction sequence changes, better modeling the semantic changes in events brought about by new user participation. To reduce the model's time complexity, the target user and users in the interaction sequence use only unique identifiers as their features and share parameters with the embedded representation layer.

[0079] There are many ways to implement attention mechanisms. This invention uses a feedforward neural network, which has greater expressive power than the attention mechanism implemented with dot products. Its calculation formula is as follows:

[0080]

[0081] Here, Concate represents vector concatenation, MLP stands for Multilayer Perceptron, and it uses the ReLU activation function.

[0082] The attention mechanism implemented in this invention does not perform normalization similar to that in the Transformer multi-head self-attention mechanism. This is because it aims to reflect the magnitude of the influence between user interaction behaviors of different lengths. For example, assuming that user interaction sequences are of different lengths, and that the users in the user interaction sequences are all friends of the user to be recommended, then the model's values ​​for longer sequences will be higher than those for shorter sequences. If attention coefficient Softmax normalization is performed, the output value will be limited to a certain range, losing some of its expressive power.

[0083] Output layer: A common multilayer perceptron-based output layer is used, which concatenates and maps all features multiple times, and finally converts them into the probability of user participation in the event through a function, as shown in the following formula:

[0084] h 0 =Concate(x interact ,x1,x2,...,x n )

[0085] h 1 =Relu(W 1 h 0 +b 1 )

[0086] h 2 =Relu(W 2 h 1 +b 2 ) ...

[0088]

[0089] Where, x n Let h be the nth feature representation vector, k be the number of layers in the multilayer perceptron, and sigmoid and ReLU are both non-linear activation functions. -1 Let be the state vector of the (k-1)th layer perceptron. W is the output vector. k Let b be the parameter of the k-th layer perceptron. k Let n be the bias vector of the k-th layer perceptron, and n be the number of feature representation vectors.

[0090] Ultimately, the deep interaction-aware event recommendation model obtains the participation prediction probability of a user sample for a specific event, which is used to score the event and make recommendations.

[0091] In one possible implementation, the model training process is as follows:

[0092] Common loss functions used in recommender systems include point-aware, pair-aware, and sequence-aware loss functions. Considering the contextual features of the interactions in this model and the scarcity of negative samples, this paper adopts a point-aware loss function and generates negative samples through negative sampling to aid model learning. Using pair-aware or sequence-aware loss functions would significantly increase the number of input features and reduce training efficiency. The model employs a bidirectional approach in generating negative samples, sampling both events and users. The loss function used during training is shown below:

[0093]

[0094] Where l is the loss, N is the number of samples, (x,y) is a single sample, D is the sample set, and y is the true value. The output vector is the predicted value.

[0095] The training process takes the training dataset, learning rate, event and user sets, and the number of samples for each event and user as input. First, the model parameters are randomly initialized. Then, the training is iterated on the training set until convergence, and SGD is used to update the model parameters.

[0096] In summary, the event recommendation method based on user interaction sequences in this invention addresses two problems existing in current event recommendation methods: (1) It is difficult to capture the semantic changes of events caused by the participation of new users, thus affecting the quality of event recommendations. (2) Event recommendations are also plagued by the cold start problem due to the timeliness and long-tail distribution of social events, resulting in a decrease in the recommendation quality of unpopular and emerging events. By introducing event user interaction sequences as features, the model's ability to model event semantics is improved. The user interaction sequence perception layer based on the attention mechanism can conduct "deep interaction" between the target user and all participating users of the event to cope with the semantic changes brought about by the participation of new users. This invention also fully explores various factors affecting user participation in events, providing rich feature inputs for the model. In addition, this invention proposes a group dependency initialization strategy to improve the recommendation effect of emerging and unpopular events.

[0097] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0098] In another embodiment of the present invention, an event recommendation system based on user interaction sequences is provided, which can be used to implement the above-mentioned event recommendation method based on user interaction sequences. Specifically, the event recommendation system based on user interaction sequences includes a data acquisition module, a model processing module, and an event recommendation module.

[0099] The data acquisition module is used to acquire users to be recommended and each event to be recommended, and to extract the recommendation influence features of users to be recommended, as well as the feature representation vectors and user interaction sequences of each event to be recommended; the model processing module is used to obtain the participation prediction probability of users to be recommended for each event to be recommended by calling a preset deep interaction perception event recommendation model based on the recommendation influence features of users to be recommended, the feature representation vectors and user interaction sequences of each event to be recommended; the event recommendation module is used to recommend each event to be recommended to users in descending order of participation prediction probability.

[0100] All relevant content of each step involved in the aforementioned embodiments of the event recommendation method based on user interaction sequences can be referenced to the functional description of the corresponding functional module of the event recommendation system based on user interaction sequences in the embodiments of the present invention, and will not be repeated here.

[0101] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0102] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of an event recommendation method based on user interaction sequences.

[0103] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the event recommendation method based on user interaction sequences in the above embodiments.

[0104] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0106] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0107] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An event recommendation method based on user interaction sequences, characterized in that, include: Obtain the users to be recommended and each event to be recommended, extract the recommendation impact features of the users to be recommended, and extract the feature representation vectors and user interaction sequences of each event to be recommended; Based on the recommendation influence characteristics of the users to be recommended, as well as the feature representation vectors of each event to be recommended and the user interaction sequence, the predicted participation probability of the users to be recommended for each event to be recommended is obtained by calling the preset deep interaction perception event recommendation model. Recommend each recommended event to the user in descending order of the probability of participating in the prediction; The deep interactive perception event recommendation model includes an embedded representation layer, a user interaction sequence perception layer based on an attention mechanism, and an output layer based on a multilayer perceptron. The embedding representation layer encodes the numerical recommendation influence features of the user to be recommended using the CLIP model to obtain the first representation vector of the user to be recommended; it encodes the categorical recommendation influence features of the user to be recommended using a pre-trained embedding vector matrix to obtain the second representation vector of the user to be recommended; the first and second representation vectors of the user to be recommended are concatenated to obtain the feature representation vector of the user to be recommended; the user interaction sequence perception layer obtains the feature representation vector of each user in the user interaction sequence of each event to be recommended, and performs interaction operations between the feature representation vector of the user to be recommended and the feature representation vector of each user in the user interaction sequence of each event to be recommended to obtain the interaction vector between the user to be recommended and each event to be recommended; the output layer obtains the participation prediction probability of the user to be recommended for each event to be recommended based on the feature representation vector of each event to be recommended and the interaction vector between the user to be recommended and each event to be recommended. The attention-based user interaction sequence perception layer is specifically as follows: in, Let be the interaction vector between the user to be recommended and the event to be recommended. Represents the attention mechanism, Let be the feature representation vector of the user to be recommended. Events awaiting recommendation e User interaction sequence i The feature representation vector, m Events awaiting recommendation e The length of the user interaction sequence, The attention coefficient, which represents the attention mechanism generated from two vectors of the same length, is a scalar. This represents a vector concatenation operation. This represents a multilayer perceptron.

2. The event recommendation method based on user interaction sequences according to claim 1, characterized in that, The extracted feature representation vectors for each event to be recommended include: Obtain the recommendation impact features of each event to be recommended; encode the numerical recommendation impact features of each event to be recommended using the CLIP model to obtain the first representation vector of each event to be recommended; encode the categorical recommendation impact features of each event to be recommended using a pre-trained embedding vector matrix to obtain the second representation vector of each event to be recommended; concatenate the first and second representation vectors of each event to obtain the recommendation impact features of each event to be recommended.

3. The event recommendation method based on user interaction sequences according to claim 2, characterized in that, The recommendation influence characteristics of the users to be recommended include one or more of the following: age, gender, city, and tag preference; the recommendation influence characteristics of the events to be recommended include one or more of the following: participating age, participating gender, number of participating users, participating city, participating week, participating distance, and tag preference.

4. The event recommendation method based on user interaction sequences according to claim 2, characterized in that, When extracting the feature representation vector of each event to be recommended, if the event to be recommended is a newly created event or a less popular event, the feature representation vector of the event to be recommended is obtained in the following way: The feature representation vector of the historical interaction events of the group to which the event to be recommended belongs is averaged using the following formula to obtain the feature representation vector of the event to be recommended: in, Let be the feature representation vector of the event to be recommended. It is an aggregate function with shift invariance. This is a collection of historical events belonging to the group to which the event to be recommended belongs. for China Event The feature representation vector.

5. The event recommendation method based on user interaction sequences according to claim 1, characterized in that, The output layer based on the multilayer perceptron is specifically: ... in, For the first n Each feature representation vector k This represents the number of layers in the multilayer perceptron. and All are non-linear activation functions. For the first k -1 layer perceptron state vector For the output vector, For the first k Parameters of the layer perceptron For the first k The bias vector of the layer perceptron. The number of feature vectors.

6. The event recommendation method based on user interaction sequences according to claim 1, characterized in that, The deep interactive event recommendation model uses the following loss function during training: in, For loss, For the number of samples, For one sample, For the sample set, For the true value, The output vector is the predicted value.

7. An event recommendation system based on user interaction sequences, characterized in that, include: The data acquisition module is used to acquire users to be recommended and each event to be recommended, extract the recommendation impact features of users to be recommended, and extract the feature representation vectors and user interaction sequences of each event to be recommended. The model processing module is used to obtain the predicted probability of the user's participation in each recommended event by calling a preset deep interaction-aware event recommendation model, based on the recommendation influence characteristics of the user to be recommended, the feature representation vector of each recommended event, and the user interaction sequence. The event recommendation module is used to recommend events to users in descending order of their probability of participating in the prediction. The deep interactive perception event recommendation model includes an embedded representation layer, a user interaction sequence perception layer based on an attention mechanism, and an output layer based on a multilayer perceptron. The embedding representation layer encodes the numerical recommendation influence features of the user to be recommended using the CLIP model to obtain the first representation vector of the user to be recommended; it encodes the categorical recommendation influence features of the user to be recommended using a pre-trained embedding vector matrix to obtain the second representation vector of the user to be recommended; the first and second representation vectors of the user to be recommended are concatenated to obtain the feature representation vector of the user to be recommended; the user interaction sequence perception layer obtains the feature representation vector of each user in the user interaction sequence of each event to be recommended, and performs interaction operations between the feature representation vector of the user to be recommended and the feature representation vector of each user in the user interaction sequence of each event to be recommended to obtain the interaction vector between the user to be recommended and each event to be recommended; the output layer obtains the participation prediction probability of the user to be recommended for each event to be recommended based on the feature representation vector of each event to be recommended and the interaction vector between the user to be recommended and each event to be recommended. The attention-based user interaction sequence perception layer is specifically as follows: in, Let be the interaction vector between the user to be recommended and the event to be recommended. Represents the attention mechanism, Let be the feature representation vector of the user to be recommended. Events awaiting recommendation e User interaction sequence i The feature representation vector, m Events awaiting recommendation e The length of the user interaction sequence, The attention coefficient, which represents the attention mechanism generated from two vectors of the same length, is a scalar. This represents a vector concatenation operation. This represents a multilayer perceptron.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the event recommendation method based on user interaction sequences as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the event recommendation method based on user interaction sequences as described in any one of claims 1 to 6.

Images