Service recommendation method and system based on interaction between location and preference features

Abstract

The present invention discloses a service recommendation method based on the interaction of location and preference features, comprising: obtaining identification information, location information, and historical call records of users and services; obtaining a service set corresponding to the user and a user set corresponding to the service based on the historical call information and location information of the user and service, respectively; fusing and extracting the interest preferences of the user and service using a multi-head attention fusion network and an interest feature extraction network; subjecting the interest preferences, location information, and identification information of the user and service to feature interaction, inputting the information into a deep residual neural network, and outputting a service quality prediction result; and selecting the service with the best service quality prediction result as the final recommended service. The present invention addresses the problems of data redundancy and imprecision by locating the interest preferences of users and services, integrating the location information and interest preferences of users and services into the prediction of service quality, and performing linear and nonlinear transformations on the features to improve the accuracy of predictions and recommendations.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent recommendation technology, and in particular relates to a service recommendation method and system based on the interaction of location and preference features. Background Art

[0002] With the rapid development of the Internet and the popularization of cloud computing and big data, people's quality of life has also improved, and as a result, people's demand for personalized recommendation services has increased. How to accurately recommend personalized services to people has become an issue worthy of study.

[0003] Quality of Service (QoS), a non-functional attribute of a service, is a key criterion for evaluating its performance and utility. Using QoS-based web service recommendation techniques to recommend the most suitable web services to users has become a research hotspot in the field of service computing in recent years. Improving the accuracy of service QoS attribute value predictions can yield more reasonable and effective web service recommendations.

[0004] At the same time, when there is less data on user and service interaction behavior and historical behavior, data sparsity can occur. This can lead to missing data and uneven sampling during analysis and recommendation, resulting in poor generalization, poor training results, and ultimately biased recommendations. Furthermore, existing methods often analyze only a single feature of the user and service. This approach is limited in its singleness and fails to fully integrate the characteristics of the user and service, nor consider their underlying information, which can ultimately lead to poor recommendations. Summary of the Invention

[0005] In order to solve the problem of low accuracy of recommendation results in existing recommendation methods, the present invention provides a service recommendation method and system based on the interaction of location and preference features. According to the interaction records and context information of users and services, the interest preferences of users and services are extracted to solve the redundancy and inaccuracy problems caused by large amounts of data, and at the same time reflect the location preferences of users and services; the interest preferences, location information and identification information of users and services are integrated and input into a neural network, which can better reflect the correlation between users and services; in order to better reflect the implicit relationship between users and services, deep feature interaction of features is performed using a neural network; accurate prediction of service quality is completed through a deep residual neural network, thereby optimizing the service recommendation results and improving the accuracy of service recommendations.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] A first aspect of the present invention provides a service recommendation method based on the interaction of location and preference features.

[0008] A service recommendation method based on the interaction of location and preference features, comprising:

[0009] Obtain user and service identification information, location information, and historical call records;

[0010] Based on the historical call information and location information of users and services, we obtain the service set corresponding to the user and the user set corresponding to the service. We use the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of users and services.

[0011] After the user and service interest preferences, location information, and identification information are interactively analyzed, they are input into a deep residual neural network to output the service quality prediction results.

[0012] The service with the best service quality prediction result is selected as the final recommended service.

[0013] According to a further technical solution, the historical call records include interaction records between users and services.

[0014] A further technical solution is to determine other services called by the user based on the user's historical call information, filter the services based on the location information, and then use the modified cosine similarity to calculate the similarity between each of the filtered services and the service, and filter again based on the similarity to obtain a service set.

[0015] A further technical solution is to determine other users who call the service based on the historical call information of the service, screen the users based on their location information, and then use the modified cosine similarity to calculate the similarity between each of the screened users and the user, and screen again based on the similarity to obtain a user set.

[0016] Further technical solutions, the similarity calculation formula between services is:

[0017]

[0018] Among them, S u,v is the similarity calculation between service u and service v, I uv is the public user set of service u and service v, I u is the set of users of service u, I v is the set of users of service v, R u,t is the QoS value generated when service u is called by user t, R v,t is the QoS value generated when service v is called by user t, is the average QoS value of service u on the user set, is the average QoS value of service v on the user set.

[0019] Further technical solution, the formula for calculating the similarity between users is:

[0020]

[0021] Among them, S a,b is the similarity between user a and user b, I ab is the public service set of user a and user b, I a is the service set of user a, I b is the service set of user b, R a,c is the QoS value generated when user a calls service c, R b,c is the QoS value generated when user b calls service c, is the average QoS value of user a on the service set, is the average QoS value of user b on the service set.

[0022] A further technical solution is to extract the interest preference features of users and services. The fused service feature group and user feature group are respectively input into a single-layer neural network to compress the information and learn the weight of each dimension. The fused features are combined with the learned weights to extract the interest preference features of users and services.

[0023] A second aspect of the present invention provides a service recommendation system based on the interaction of location and preference features.

[0024] A service recommendation system based on the interaction of location and preference features, comprising:

[0025] Data information acquisition module, used to obtain user and service identification information, location information and historical call records;

[0026] The data processing module is used to obtain the service set corresponding to the user and the user set corresponding to the service based on the historical call information and location information of the user and the service, and use the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of the user and the service;

[0027] The service quality prediction module is used to perform feature interaction between the user and service interest preferences, location information, and identification information, input them into a deep residual neural network, and output the service quality prediction results;

[0028] The service recommendation module is used to select the service with the best service quality prediction result as the final recommended service.

[0029] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor. When the computer instructions are executed by the processor, the steps of the service recommendation method based on the interaction between location and preference features are completed.

[0030] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executed by the above-mentioned service recommendation method based on the interaction between location and preference features.

[0031] One or more of the above technical solutions have the following beneficial effects:

[0032] (1) The present invention extracts the interests and preferences of users and services based on the interaction records of users and services and context information. It fully utilizes the context information and the interaction records of users and services to better locate the interests and preferences of users and services, reduce the number and proportion of special samples, and solve the problems of data redundancy and inaccuracy.

[0033] (2) The present invention combines the interest preferences, location information and identification information of users and services as inputs for neural network training and learning. In order to better reflect the interactive relationship between users and services, the neural network is used to perform feature interaction on features to complete the final prediction and recommendation; the location information and interest preferences of users and services are integrated, and the neural network is used to perform linear and nonlinear transformations to improve the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0035] Figure 1 This is a flowchart of the service recommendation method based on the interaction of location and preference features according to the first embodiment of the present invention;

[0036] Figure 2 This is a theoretical framework diagram of the service recommendation method based on the interaction between location and preference features according to the first embodiment of the present invention;

[0037] Figure 3 Schematic diagram of a method for extracting user and service interest preference features in Embodiment 1 of the present invention;

[0038] Figure 4 This is a schematic diagram of the feature interaction and QoS prediction method in Example 1 of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] Example 1

[0042] In order to solve the problem of low accuracy of recommendation results in existing recommendation methods, the present invention provides a service recommendation method and system based on the interaction of location and preference features. According to the interaction records and context information of users and services, the interest preferences of users and services are extracted to solve the problems of data redundancy and inaccuracy. The interest preferences are then combined with the location information and identification information of users and services to perform feature interaction, and accurate prediction of service quality QoS is completed through a deep residual neural network, thereby optimizing the service recommendation results and improving the accuracy of service recommendations.

[0043] like Figure 1 As shown, this embodiment discloses a service recommendation method based on the interaction of location and preference features, including:

[0044] Obtain user and service identification information, location information, and historical call records;

[0045] Based on the historical call information and location information of users and services, we obtain the service set corresponding to the user and the user set corresponding to the service. We use the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of users and services.

[0046] After the user and service interest preferences, location information, and identification information are interactively analyzed, they are input into a deep residual neural network to output the service quality prediction results.

[0047] The service with the best service quality prediction result is selected as the final recommended service.

[0048] In this embodiment, the theoretical framework of the service recommendation method based on the interaction of location and preference features is shown in FIG. Figure 2 shown.

[0049] First, the identification information, location information and historical call records of the user and service are obtained, wherein the identification information refers to an attribute of the user and service, the ID and identification information of the user and service.

[0050] The historical call records are interaction records between users and services. These interaction records are also known as QoS records. The underlying characteristics of QoS can reflect the interests and preferences of users and services, and have a certain impact on the recommendation process. However, QoS attributes are very susceptible to the influence of user and service context information. Combining the location information of users and services, similar users and similar services are calculated for users and services. These similar users and similar services are used to filter out user-specific service feature vector groups and service-specific user feature vector groups, solving the redundancy and inaccuracy problems caused by large amounts of data while reflecting the location preferences of users and services.

[0051] The context information of users and services refers to the context of users and services. This information includes multiple contents. In this embodiment, it refers to the location information of users and services.

[0052] Then, through the multi-head attention and interest extraction network, the interest preferences of users and services are extracted from both user and service aspects, increasing the probability of samples appearing in a large range and reducing the proportion of individual samples, which is in line with the interest preferences of users and services.

[0053] Finally, the extracted user and service interest preferences are combined with location information and identification information as input to the residual network for training and learning. This in-depth study of personalized recommendations from three perspectives, location, interest, and identification information, is conducted. To better reflect the implicit relationship between users and services, vector features are linearly interacted, allowing deeper networks to better capture the hidden characteristics of the data. However, as the network deepens, it also introduces the problem of network degradation. Therefore, the introduction of ResNet into the neural network enables the network to better learn nonlinear features. By optimizing the neural network through feature reuse, more accurate prediction results can be achieved as the network depth increases.

[0054] In this embodiment, if Figure 3 As shown, first, the identification information, location information and historical call records of users and services are obtained, and based on the historical call records and location information of users and services, the service set corresponding to the user and the user set corresponding to the service are obtained respectively. The multi-head attention fusion network and interest feature extraction network are used to fuse and extract the interest preferences of users and services.

[0055] Among them, historical call records are interaction records between users and services. Contextual information has a certain impact on user recommendations. Services or users with similar locations may have similar interests and preferences. Therefore, the historical call information of users and services is filtered in combination with the context of users and services to select effective QoS records.

[0056] Take the QoS prediction when user Z calls a service z as an example. Assume that the set of other services called by user Z is S={S1, S2, ... S n}, the set of users whose service z has been called by other users is U = {U1, U2, ... U n}, and perform similarity calculations on the two aspects of user Z and service z respectively.

[0057] Specifically, for the user, for the QoS prediction when user Z calls a service z, the other service set S called by user Z is determined based on the historical call records of user Z = {S1, S2, ... S n}, select services in the same country and the same AS (Autonomous System) as service z to form a new service set S′, use the modified cosine similarity to calculate the similarity between each service in the selected service set S′ and service z, and take the top k similar services in descending order of similarity to form a new service set S″.

[0058] The formula for calculating the similarity between services is:

[0059]

[0060] Among them, S u,v is the similarity calculation between service u and service v, I uv is the public user set of service u and service v, I u is the set of users of service u, I v is the set of users of service v, R u,t is the QoS value generated when service u is called by user t, R v,t is the QoS value generated when service v is called by user t, is the average QoS value of service u on the user set, is the average QoS value of service v on the user set.

[0061] If the number of services in the service set S that are in the same country and the same AS as service z is too small, less than k, and assuming it is j, then expand the screening range and select services in the service set S except for the service set S′ that are in the same country as service z. Use the modified cosine similarity to calculate the similarity, take the first kj similar services, and form a total of k similar services with the j services in the service set S′, thereby forming a new service set S″.

[0062] If the total number of services in the service set S is very small, less than k, the similarity between services is no longer calculated, but the service set S is directly determined for subsequent calculations.

[0063] In terms of services, the similar users of a user are similar to the above scheme. Based on the historical call information of service z, that is, the historical information of service z being called by other users, the user set U = {U1, U2, ... U n}, filter out the user set U′ that is in the same country and the same AS as user Z, calculate the similarity between each user in the filtered user set U′ and user Z using the modified cosine similarity, and take the top k similar users in descending order of similarity to form a new user set U″.

[0064] The formula for calculating the similarity between users is:

[0065]

[0066] Among them, S a,b is the similarity between user a and user b, I ab is the public service set of user a and user b, I a is the service set of user a, I b is the service set of user b, R a,c is the QoS value generated when user a calls service c, R b,c is the QoS value generated when user b calls service c, is the average QoS value of user a on the service set, is the average QoS value of user b on the service set.

[0067] If the number of users in user set U that are in the same country and the same AS as user Z is too small, less than k, and assuming it is j, then expand the screening range and select users in user set U excluding user set U′ that are in the same country as user Z. Use the modified cosine similarity to calculate the similarity, take the first kj similar users, and form a total of k similar users with the j users in user set U′ to form a new user set U″.

[0068] If the total number of users in the user set U is very small, less than k, the similarity between users is no longer calculated, but the user set U is directly determined for subsequent calculations.

[0069] Through the above scheme, similar users and similar services are calculated to obtain a set of users with consistent preferences U″={U1,U2,…U k} or service set S″={S1,S2,…S k}, respectively obtain their QoS, that is, QoS records, to form user feature group X U ={x u1 , x u2 ,…x uk} and service feature group X I ={x i1 , x i2 ,…x ik}, using multi-head attention fusion network and interest feature extraction network to fuse and extract interest preferences of both users and services.

[0070] Although the preferences of the obtained user feature group and service feature group are consistent, there may be differences between services or between users. There may also be deviations in the calls of different users to the same service and the calls of the same user to different services. In order to solve the problem of large deviations and avoid the impact of deviations, and to extract more accurate interest preference features, this embodiment uses a multi-head attention mechanism to fuse the user feature group and the service feature group.

[0071] The multi-head attention fusion network uses the multi-head attention mechanism to perform multiple groups of attention processing on user features and service features to obtain fused features. Each feature in the fused features can be explained by the feature itself and other related features.

[0072] Assume that the user feature group is X U ={x u1 , x u2 ,…x uk}, service feature group is X I ={x i1 , x i2 ,…x ik}, using the multi-head attention mechanism to fuse features.

[0073] First, to integrate user interest preference features, the service feature group is compressed into a low-dimensional, dense vector group using an embedding layer. A single attention process is performed first, and the similarity between each vector in the vector group and all other vectors is calculated during the process. To avoid excessive inner products during training, a scaling factor is added to adjust the size of the inner product. Then, normalization is performed to obtain a weight distribution, and the vectors are weighted. The specific formula is:

[0074]

[0075] Among them, Q, K, and V are the query matrix, key matrix, and value matrix. The query, key, and value of the eigenvector are obtained through linear mapping. is the scaling factor, d k It is the dimensional feature of Q and K, and softmax is the activation function.

[0076] Use different trainable matrices to linearly map the query, key, and value h times. Each scaling dot product is a head, and h times means h heads. Parameters are not shared. The expression formula is as follows:

[0077]

[0078] in, is the trainable matrix of the jth linear mapping, head j is the jth head. The result of h times of scaling dot product operation is concatenated and then linearly transformed to obtain the value as the result of multi-head attention. The formula is:

[0079] MHAttention(X u )=Concat(head1,...head h )W O

[0080] Among them, W o is a learnable parameter matrix used to calculate the correlation between different heads.

[0081] Through the above multi-head attention fusion network, input the service feature group X I ={x i1 , x i2 ,…x ik}Fuse features and output the fused feature group containing k features.

[0082] Similar to the fusion of interest preference features in service, user feature group X U ={x u1 , x u2 ,…x ukInput the above multi-head attention fusion network to perform feature fusion, and output the fused feature group containing k features.

[0083] Information fusion is achieved by comparing the similarities and weights between vectors in different subspaces through multi-head attention operations. The fused information is encoded and each vector can be represented by itself and others, which enables the information between vectors to be fused and complement each other.

[0084] Based on the fused features, the interest feature extraction network is used to extract interest preference features.

[0085] To extract user interest preference features, the fused service feature group is input into a single-layer neural network to compress information and learn the weights of each dimension. The fused features are combined with the learned weights to extract user interest preference features. The formula is as follows:

[0086] U P =X' I ·f(X' I )=X' I ·σ1(W1·X' I +b1)

[0087] Among them, U P is the user interest preference feature, X' I is the output of the fusion network, W1 is the weight matrix, b1 is the bias, σ1 is the activation function, and the activation function selects the softmax function.

[0088] The extraction of service interest preference features is similar. The fused user feature group is input into a single-layer neural network to compress information and learn the weight of each dimension. The fused features are combined with the learned weights to extract service interest preference features.

[0089] The above steps complete the extraction of interest preference features of both users and services.

[0090] like Figure 4 As shown in the figure, based on the extracted user interest preferences and service interest preferences, combined with the identification information and location information of the user and service, feature interaction is performed on the user features and service features, and the deep residual neural network is used to complete the service quality prediction.

[0091] Specifically, the embedding vectors of user identification information and user location information are connected with the user's interest preference features to form a one-dimensional vector, namely the user feature vector. During the prediction process, the correlation of various features is fully considered to improve the accuracy of the prediction. The formula for the user feature vector is as follows:

[0092]

[0093] Among them, U I is the user identification information, U L is the user location information, U P is the user's interest preference feature, x U It is a user feature.

[0094] Similarly, the service feature vector calculation process is similar. Specifically, the embedding vectors of the service identification information and service location information are connected with the service interest preference features to form a one-dimensional vector, namely the service feature vector. During the prediction process, the correlation of various features is fully considered to improve the accuracy of the prediction. The service feature vector formula is as follows:

[0095]

[0096] Among them, I I Is the service identification information, I L is the service location information, I P is the interest preference feature of the service, x I It is a service feature.

[0097] For user feature vectors and service feature vectors, since the dimensions of the feature vectors of users and services are the same, feature interaction is completed by multiplying the elements in the same position between the vectors to achieve linear changes in features. The formula is as follows:

[0098] v=Φ(x I ,x U )=x I ⊙x U

[0099] Deep residual neural networks are constructed using residual connections. Residual connections reuse features to address network degradation as the network grows deeper, enabling nonlinear feature changes. This improves network performance and prediction accuracy in deep neural networks. The neural network consists of multiple residual blocks, each consisting of a main path and a shortcut. The shortcut adds the original vector to the main path to achieve feature reuse.

[0100] Use a fully connected layer to map v in the shortcut i , so that it can be successfully added to the main path, W i T represents the weight matrix, b i Indicates bias:

[0101] n i =W i T v i +b i

[0102] Before the features on the main path pass through the first and second fully connected layers, W i T represents the weight matrix, b i Indicates bias:

[0103] m i =W i T f(W i T v i +b i )+b i

[0104] Before using the activation function in the second fully connected layer, add the vector on the shortcut, and then output it after activation:

[0105] v i+1 =f(s i )=f(n i +m i )

[0106] The output layer of the deep residual neural network is mainly used to output the final prediction result. The output layer is fully connected to the output of the last residual block and is used for prediction:

[0107]

[0108] in, Represented as the prediction result, Represented as the weight matrix of the output layer, b h Denotes the bias of the output layer, v n is the output of the last residual block, and σ is the activation function.

[0109] Finally, the service with the best service quality prediction result is selected as the final recommended service. Specifically, through the above scheme, a service quality prediction result relative to the user is calculated. This prediction result is more accurate than existing calculation schemes. Based on this prediction result, the service with the best prediction result is selected as the final recommended service, achieving optimization and improvement of service recommendations and improving the accuracy of recommendations.

[0110] Example 2

[0111] This embodiment discloses a service recommendation system based on the interaction of location and preference features, including:

[0112] Data information acquisition module, used to obtain user and service identification information, location information and historical call records;

[0113] The data processing module is used to obtain the service set corresponding to the user and the user set corresponding to the service based on the historical call information and location information of the user and the service, and use the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of the user and the service;

[0114] The service quality prediction module is used to perform feature interaction between the user and service interest preferences, location information, and identification information, input them into a deep residual neural network, and output the service quality prediction results;

[0115] The service recommendation module is used to select the service with the best service quality prediction result as the final recommended service.

[0116] Example 3

[0117] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the service recommendation method based on the interaction between location and preference features are completed.

[0118] Example 4

[0119] This embodiment further provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the service recommendation method based on the interaction between location and preference features are completed as described above.

[0120] The steps involved in the above embodiments 2 to 4 correspond to those in the method embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0121] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0122] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

[0123] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

[0124] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A service recommendation method based on the interaction of location and preference features, characterized in that: include: Obtain user and service identification information, location information, and historical call records; Based on the historical call information and location information of the user and the service, a service set corresponding to the user and a user set corresponding to the service are obtained respectively. Specifically, the following steps are performed: based on the historical call information of the user, other services called by the user are determined; after filtering the services based on the location information, the similarity between each of the filtered services and the service is calculated using the modified cosine similarity; and the similarity is filtered again based on the similarity to obtain a service set; based on the historical call information of the service, other users who called the service are determined; after filtering the users based on the location information, the similarity between each of the filtered users and the user is calculated using the modified cosine similarity; and the similarity is filtered again based on the similarity to obtain a user set; Utilize the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of users and services; The interest preferences, location information, and identification information of users and services are subjected to feature interaction. Specifically, based on the extracted user interest preferences and service interest preferences, combined with the identification information and location information of the user and service, feature interaction is performed on the user features and service features. Specifically, the embedding vectors of the user identification information and user location information are connected with the user interest preference features to form a one-dimensional vector, namely the user feature vector. In the prediction process, the correlation of various features is fully considered to improve the accuracy of the prediction. The formula of the user feature vector is as follows: in, is user identification information, is the user's location information, is the user's interest preference feature, is a user feature; Similarly, the service feature vector calculation process is similar; specifically, the embedding vectors of the service identification information and service location information are connected with the service interest preference feature to form a one-dimensional vector, namely the service feature vector In the prediction process, the correlation of various features is fully considered to improve the accuracy of the prediction; the service feature vector formula is as follows: in, Is the service identification information, is the service location information, is the interest preference feature of the service, It is a service characteristic; For user feature vectors and service feature vectors, since the dimensions of the feature vectors of users and services are the same, feature interaction is completed by multiplying the elements in the same position between the vectors to achieve linear changes in features. The formula is as follows: ； in, It is the feature vector obtained by the feature interaction between the user feature vector and the service feature vector; After that, they are fed into a deep residual neural network to output the service quality prediction results; The service with the best service quality prediction result is selected as the final recommended service.

2. The service recommendation method based on the interaction of location and preference features as claimed in claim 1, characterized in that: The historical call records include interaction records and context information between users and services.

3. The service recommendation method based on the interaction of location and preference features as claimed in claim 1, characterized in that: The formula for calculating the similarity between services is: in, is the similarity calculation between service u and service v, is the set of common users of services u and v, is the set of users of service u, is the set of users of service v, is the QoS value generated when service u is called by user t, is the QoS value generated when service v is called by user t, is the average QoS value of service u on the user set, is the average QoS value of service v on the user set.

4. The service recommendation method based on the interaction of location and preference features as claimed in claim 1, characterized in that: The formula for calculating the similarity between users is: in, is the similarity between user a and user b, is the public service set of user a and user b, is the service set of user a, is the service set of user b, is the QoS value generated when user a calls service c, is the QoS value generated when user b calls service c, is the average QoS value of user a on the service set, is the average QoS value of user b on the service set.

5. The service recommendation method based on the interaction of location and preference features as claimed in claim 1, characterized in that: In order to extract the interest preference features of users and services, the fused service feature group and user feature group are respectively input into a single-layer neural network to compress the information and learn the weight of each dimension. The fused features are combined with the learned weights to extract the user and service interest preference features.

6. A service recommendation system based on the interaction of location and preference features, characterized by: include: Data information acquisition module, used to obtain user and service identification information, location information and historical call records; The data processing module is used to obtain a service set corresponding to the user and a user set corresponding to the service based on the historical call information and location information of the user and the service, specifically: determining other services called by the user based on the historical call information of the user, filtering the services based on the location information, and then calculating the similarity between each of the filtered services and the service using the modified cosine similarity, and filtering again based on the similarity to obtain the service set; determining other users who called the service based on the historical call information of the service, filtering the users based on the location information, and then calculating the similarity between each of the filtered users and the user using the modified cosine similarity, and filtering again based on the similarity to obtain the user set; Utilize the multi-head attention fusion network and interest feature extraction network to fuse and extract the interest preferences of users and services; The service quality prediction module is used to perform feature interaction between the user and service interest preferences, location information, and identification information. Specifically, based on the extracted user interest preferences and service interest preferences, combined with the user and service identification information and location information, feature interaction is performed on the user features and service features. Specifically, the embedding vectors of the user identification information and user location information are connected with the user interest preference features to form a one-dimensional vector, namely the user feature vector. In the prediction process, the correlation of various features is fully considered to improve the prediction accuracy. The formula for the user feature vector is as follows: in, is user identification information, is the user's location information, is the user's interest preference feature, is a user characteristic; Similarly, the service feature vector calculation process is similar. Specifically, the embedding vectors of service identification information and service location information are connected with the service interest preference features to form a one-dimensional vector, namely the service feature vector. During the prediction process, the correlation of various features is fully considered to improve the prediction accuracy. The service feature vector formula is as follows: in, Is the service identification information, is the service location information, is the interest preference feature of the service, It is a service characteristic; For user feature vectors and service feature vectors, since the dimensions of the feature vectors of users and services are the same, feature interaction is completed by multiplying the elements in the same position between the vectors to achieve linear changes in features. The formula is as follows: ； in, It is the feature vector obtained by the feature interaction between the user feature vector and the service feature vector; Then input the deep residual neural network and output the service quality prediction result; The service recommendation module is used to select the service with the best service quality prediction result as the final recommended service.

7. An electronic device, characterized in that: The invention comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the service recommendation method based on the interaction of location and preference features as claimed in any one of claims 1 to 5 are completed.

8. A computer-readable storage medium, characterized in that: There are multiple instructions, which are suitable for being loaded by a processor of a terminal device and executed by a service recommendation method based on the interaction of location and preference features according to any one of claims 1 to 5.

Images