Information service system based on big data and control method

Abstract

The invention discloses an information service system based on big data and a control method, and relates to the technical field of information, and the method comprises the steps: collecting multi-modal data; based on the collected multi-modal data, multi-modal feature vectors are extracted, weight distribution is carried out on multi-modal features, and a weighted multi-modal feature set is generated; performing feature fusion based on the weighted multi-modal feature set to generate a user portrait vector; according to the user portrait vector, adopting a reinforcement learning model to generate personalized information service optimal recommendation content; and performing dynamic pushing based on the personalized information service optimal recommendation content. According to the method, through deep fusion of the user portraits and introduction of a reinforcement learning algorithm, the problem that recommendation content is lack of real-time feedback optimization in the prior art is solved, and finally the accuracy, adaptability and intelligent level of information service are improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an information service system and control method based on big data. Background Art

[0002] In recent years, with the rapid development of Internet technology and the wide application of big data, information service systems have played an increasingly important role in various fields of society. Modern information service systems provide users with services such as personalized content recommendation, precise advertisement placement, and intelligent information push through data analysis and processing. Modern information service systems rely on the collaborative work of core technical modules such as data collection, feature extraction, user portrait generation, and recommendation algorithms. Especially in the processing of multimodal data, they can make full use of the fusion analysis of various data modalities such as text, image, video, and voice, greatly improving the comprehensiveness and personalization level of information services. Traditional information service systems mainly rely on single-modal data forms for processing, such as only analyzing the user's text input or click behavior. However, with the diversification of data sources and the complexity of user needs, single-modal technologies gradually show limitations and are difficult to meet the needs of users for various types of information services. Therefore, in recent years, the technologies of multimodal data collection and fusion have received great attention, and the introduction of artificial intelligence technologies such as reinforcement learning has also brought a new round of technological innovation to the field of information services.

[0003] However, the existing multimodal information service systems and control methods still have obvious deficiencies, mainly reflected in the following two aspects. First, the existing technologies have limitations in the extraction and fusion of multimodal features, lacking an effective mechanism for deep modeling of the internal relationships between different modal data and weight allocation, often resulting in the inability of the feature fusion results to accurately represent the true needs of users. For example, in the existing technologies, simple weighted average or linear combination methods are often used to fuse multimodal features, without fully considering the non-linear correlation between modalities and the dynamic changes of user preferences, thus affecting the accuracy of user portraits. Second, the existing information recommendation algorithms usually rely on fixed rules or traditional machine learning models to generate recommended content, lacking the ability to dynamically adjust according to the real-time behavior feedback of users, resulting in poor personalization and adaptability of the recommended results. For example, traditional recommendation systems fail to effectively use reinforcement learning models to model the long-term benefits of user-content interactions when optimizing recommendation strategies, resulting in the inability of the recommended content to be adjusted in a timely manner as user preferences change, affecting the user experience. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an information service control method based on big data to solve the problems of insufficient deep modeling and fusion of multi-modal features and lack of dynamic feedback optimization in the recommendation algorithm.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an information service control method based on big data, which includes collecting multi-modal data; extracting multi-modal feature vectors based on the collected multi-modal data and assigning weights to the multi-modal features to generate a weighted multi-modal feature set; performing feature fusion based on the weighted multi-modal feature set to generate a user profile vector; using a reinforcement learning model according to the user profile vector to generate the optimal recommended content of personalized information service; and performing dynamic push based on the optimal recommended content of personalized information service.

[0007] As a preferred solution of the information service control method based on big data according to the present invention, wherein: the multi-modal data includes text data, image data, voice data, and environmental data.

[0008] As a preferred solution of the information service control method based on big data according to the present invention, wherein: the step of extracting multi-modal feature vectors based on the collected multi-modal data is as follows, Performing word segmentation, removing stop words on the multi-modal data, and extracting semantic features to generate a text feature vector; Using a pre-trained convolutional neural network to extract an image feature vector; Converting the voice data into text and combining with an emotion analysis model to extract user emotion features to generate a voice feature vector; Performing data normalization processing on the environmental data and extracting an environmental feature vector.

[0009] As a preferred solution of the information service control method based on big data according to the present invention, wherein: the step of assigning weights to the multi-modal features to generate a weighted multi-modal feature set is as follows, Based on the text feature vector, image feature vector, voice feature vector, and environmental feature vector, dynamically calculate the weight of each modality, and the expression is: ; Wherein, represents the weight of the th modality, represents the historical interaction weight of the th modality, represents the natural exponential function, represents the index of the text, image, voice, and environmental modalities; The obtained modality weight vector Applied to the corresponding modal feature vectors to generate a weighted multi-modal feature set, the expression is: ; Among them, represents the weighted multi-modal feature set, represents the th modal feature vector.

[0010] As a preferred solution of the information service control method based on big data according to the present invention, wherein: based on the weighted multi-modal feature set, feature fusion is performed to generate a user portrait vector, and the specific steps are as follows. Based on the weighted multi-modal feature set, using the Transformer model, calculate the fusion feature representation of each modality, and the expression is: ; Among them, represents the th modal fusion feature representation, represents the th modal and the th modal feature similarity, represents the th modal and the th modal attention score, represents the th modal and the th modal attention score, represents the th modal feature vector; Concatenate the fusion feature representations of all modalities, and generate a unified user portrait vector through a fully connected layer and a non-linear activation function. The expression is: ; Among them, represents the user portrait vector, represents the activation function, represents the weight matrix of the fully connected layer, represents the bias term, represents the fusion feature of the text modality, represents the fusion feature of the image modality, represents the fusion feature of the voice modality, represents the fusion feature of the environment modality.

[0011] As a preferred solution of the information service control method based on big data according to the present invention, wherein: according to the user portrait vector, using a reinforcement learning model, generate the optimal recommended content of personalized information service, and the specific steps are as follows. The generated user portrait vector is decomposed into a multi-dimensional feature subspace, which is divided into the static feature subspace of the user and the dynamic feature subspace ; Send and into the feature embedding layer respectively to generate the state of the reinforcement learning model. The expression is: ; Among them, represents the state representation of reinforcement learning, represents the embedding matrix of static features, represents the embedding matrix of dynamic features; Based on the state representation of reinforcement learning, use the DQN algorithm to initialize the action value function, and the optimization goal is to maximize the action value. The expression is: ; Among them, represents the current state and the action value of the executed action , represents the learning rate, represents the immediate reward value, represents the discount factor, represents the future state of the optimal action corresponding to the maximum action value, represents the current action, represents the future state, represents the next state one of all actions in; According to the optimized action value function, select the action with the maximum action value as the recommended content. The expression is: ; Among them, represents the maximum action, represents the current state and the action value of the executed action , represents the candidate set of recommended content; Get the maximum action , as the optimal recommended content for personalized information services.

[0012] As a preferred solution of the information service control method based on big data described in the present invention, among them: based on the optimal recommended content of personalized information services, perform dynamic push. The specific steps are as follows, Based on the optimal recommended content of personalized information service, an optimization model is used to select the optimal modality, and the expression is: ; Among them, represents the selected optimal modality, represents the set of all available modalities, represents the preference weight for modality in the user profile, represents the adaptability score of the recommended content on modality ; represents the adaptability factor of the user's current context to modality ; According to the selected optimal modality, the optimal recommended content of personalized information service is converted into display forms of text modality, image modality, voice modality and video modality, and the push method is dynamically adjusted for pushing according to the user's device type, network condition and push timeliness.

[0013] In the second aspect, the present invention provides an information service system based on big data, including a data collection module, a weighted multi-modal feature generation module, a user profile generation module, a recommended content module and a dynamic push module; the data collection module is used to collect multi-modal data; the weighted multi-modal feature generation module is used to extract multi-modal feature vectors and assign weights to the multi-modal features based on the collected multi-modal data to generate a weighted multi-modal feature set; the user profile generation module is used to perform feature fusion based on the weighted multi-modal feature set to generate a user profile vector; the recommended content module is used to generate the optimal recommended content of personalized information service according to the user profile vector by using a reinforcement learning model; the dynamic push module is used to perform dynamic push based on the optimal recommended content of personalized information service.

[0014] In the third aspect, the present invention provides a computer device, including a memory and a processor, and the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the information service control method based on big data as described in the first aspect of the present invention is implemented.

[0015] In the fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, any step of the information service control method based on big data as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: By collecting multi-modal information such as text, images, voice, and environmental data, extracting multi-modal feature vectors and dynamically assigning weights, a weighted multi-modal feature set is generated; Subsequently, the Transformer model is used to deeply fuse the multi-modal features to generate a high-precision user portrait vector, so as to comprehensively and accurately characterize the user's interests, behavior preferences, and current environmental status; Combining with the reinforcement learning model, based on the user portrait vector, the optimal recommended content of personalized information services is dynamically generated and displayed through an optimized dynamic push method. Through the generation of weighted multi-modal features, the dynamic modeling of the non-linear relationship between modalities is realized, and the accuracy of feature fusion is enhanced; By introducing the deep fusion of user portraits and the reinforcement learning algorithm, the problem of lack of real-time feedback optimization in the recommended content in the prior art is solved, and finally the accuracy, adaptability, and intelligent level of information services are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the information service control method based on big data in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of the information service system based on big data in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the above-mentioned objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0021] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or selectively exclusive embodiment with other embodiments.

[0023] Embodiment 1, refer to Figure 1 andFigure 2 , which is the first embodiment of the present invention. This embodiment provides an information service control method based on big data, including the following steps: S1. Collect multimodal data; The multimodal data includes text data, image data, voice data, and environmental data; It should be noted that the text data provides semantic information, the image data provides visual information, the voice data provides emotional information, and the environmental data provides context information.

[0024] S2. Based on the collected multimodal data, extract multimodal feature vectors; Perform word segmentation, remove stop words on the basis of the multimodal data, and extract semantic features to generate text feature vectors; It should be noted that word segmentation divides the text into independent semantic units, removing stop words eliminates the interference of irrelevant words, and semantic feature extraction captures the deep semantics and context relationships of the text through methods such as TF-IDF and word vector models.

[0025] Use a pre-trained convolutional neural network to extract image feature vectors; It should be noted that during the extraction process, the image undergoes multiple convolutional and pooling operations, gradually extracting feature representations with deep semantics, and finally generating fixed-length image feature vectors through fully connected layers or global average pooling.

[0026] Convert the voice data into text, and combine it with an emotion analysis model to extract user emotion features to generate voice feature vectors; It should be noted that convert the voice data into a processable text form to extract semantic information; at the same time, combine it with an emotion analysis model to process features such as pitch, speech rate, volume, and rhythm in the voice signal to identify the user's emotional state.

[0027] Perform data standardization processing on the environmental data to extract environmental feature vectors; It should be noted that perform data standardization processing on environmental data such as geographical location, time, weather, device type, network status, etc., to eliminate the dimensional differences of different data dimensions and enable feature extraction on the same scale.

[0028] S3. Assign weights to the multimodal features to generate a weighted multimodal feature set; Based on the text feature vectors, image feature vectors, voice feature vectors, and environmental feature vectors, dynamically calculate the weight of each modality. The expression is: ; Where represents the weight of the th modality, represents the The historical interaction weight of a modality Denotes the natural exponential function Denotes the indices for text, image, speech, and environmental modalities; It should be noted that Denotes the natural exponential function, which is used to extend the range of weight changes to a non-linear space, making the contribution of higher historical interaction weights to the final weight more significant. By normalizing the denominator to the sum of the exponential weights of all modalities, it is ensured that the sum of the weights of the four modalities is 1.

[0029] The obtained modality weight vector Is applied to the corresponding modality feature vector to generate a weighted multi-modal feature set, and the expression is: ; Where Denotes the weighted multi-modal feature set Denotes the th modality's feature vector; It should be noted that by weighted summing the feature vectors of each modality according to their weight ratios, the importance of different modalities in multi-modal fusion is comprehensively considered, and a unified feature representation containing semantic, visual, emotional, and environmental information is generated . This feature set effectively integrates multi-modal information, provides high-quality input for subsequent personalized recommendation, classification, or user portrait generation, and dynamically adapts to the user's historical interaction behavior and current context information.

[0030] S4. Based on the weighted multi-modal feature set, perform feature fusion to generate a user portrait vector; Based on the weighted multi-modal feature set, use the Transformer model to calculate the fused feature representation of each modality, and the expression is: ; Where Denotes the fused feature representation of the th modality Denotes the feature similarity between the th modality and the th modality Denotes the attention score between the th modality and the th modality Denotes the attention score between the th modality and the th modality Denotes the feature vector of the th modality; It should be noted that Normalize the attention scores between all modalities to ensure that the sum of weights is 1, so as to dynamically adjust the contribution of each modality to . Denote the attention score between the -th modality and the -th modality, which reflects the correlation between modalities and assigns weights through normalization. By dynamically weighting the information of other modalities, fuse its own characteristics and cross-modal associations to generate a feature representation containing multi-modal interaction information, providing an optimized input for multi-modal tasks.

[0031] Concatenate the fused feature representations of all modalities and generate a unified user portrait vector through a fully connected layer and a non-linear activation function. The expression is as follows: ; wherein, denotes the user portrait vector, denotes the activation function, denotes the weight matrix of the fully connected layer, denotes the bias term, denotes the fused feature of the text modality, denotes the fused feature of the image modality, denotes the fused feature of the voice modality, denotes the fused feature of the environmental modality; It should be noted that contains the text features after interacting with other modalities, combines the image information and the interaction information of other modalities, describes the voice information and its relationship with other modalities, integrates the environmental features and the connections of other modalities, Stack the fused features of all modalities along the dimension to form a comprehensive multi-modal feature representation.

[0032] S5. According to the user portrait vector, adopt a reinforcement learning model to generate the optimal recommended content of personalized information services; Decompose the generated user portrait vector into a multi-dimensional feature subspace, which is divided into the static feature subspace of the user and the dynamic feature subspace ; It should be noted that denotes the static feature subspace of the user, which contains the long-term attributes of the user, such as fixed features like gender, age, interest preferences, etc., denotes the dynamic feature subspace, which captures the behavior, state and context information of the user at a specific time or in a specific scenario, such as real-time click behavior, current environment, etc.

[0033] Send and into the feature embedding layer respectively to generate the state of the reinforcement learning model. The expression is: ; Wherein, represents the state representation of reinforcement learning, represents the embedding matrix of static features, represents the embedding matrix of dynamic features; It should be noted that is the embedding matrix of static features, which is used to convert the static feature subspace into the state representation of reinforcement learning, and learn the weights and importance of static features in different scenarios. is the embedding matrix of dynamic features, which is used to convert the dynamic feature subspace into the state representation, and capture the influence of dynamic features in the current scenario. Both of these matrices are learnable parameters and are continuously optimized through the training process of reinforcement learning to generate a more accurate state representation.

[0034] Based on the state representation of reinforcement learning, use the DQN algorithm to initialize the action value function. The optimization goal is to maximize the action value. The expression is: ; Wherein, represents the current state and the action value of the executed action , represents the learning rate, represents the immediate reward value, represents the discount factor, represents the optimal action of the future state corresponding to the maximum action value, represents the current action, represents the future state, represents the next state one of all actions under; It should be noted that the formula updates the current action value according to the immediate rewardand the future optimal action value. This update method enables the model to gradually learn the optimal action value function, thereby guiding the agent to select the optimal action. Through the discount factor , the formula can balance the immediate reward and the future reward​, this mechanism enables the agent to focus on both short-term and long-term benefits. As the training progresses, the agent gradually approaches the optimal action-value function by continuously updating .

[0035] According to the optimized action-value function, select the action with the maximum action value as the recommended content, and the expression is: ; where, represents the maximum action, represents the current state and the action value of performing the action , represents the candidate set of the recommended content; It should be noted that <� usually corresponds to the specific content recommended to the user, such as product, video, and article recommendation items. The action value is an index in the reinforcement learning model to measure the quality of each action, and it represents the expected value of the cumulative reward that the agent can obtain after performing the action . The essence of the operation is to compare the values of all candidate actions and find the action with the highest value. represents selecting from the candidate set the action that maximizes . .

[0036] The obtained maximum action will be used as the optimal recommended content for personalized information services; It should be noted that the action is considered to be the content that the user is currently most interested in and most likely to accept, and it is also the recommended item with the greatest benefit.

[0037] S6. Based on the optimal recommended content for personalized information services, perform dynamic push; Based on the optimal recommended content for personalized information services, use the optimization model to select the optimal modality, and the expression is: ; where, represents the selected optimal modality, represents the set of all available modalities, represents the preference weight of the modality in the user profile, represents the adaptability score of the recommended content on the modality , represents the adaptability factor of the user's current context to the modality ; It should be noted that the optimal modality It can maximize the score in the formula to ensure that the recommended content can better meet the user's needs in this modality. Let \(\mathcal{M}\) denote the set of all available modalities, including image modality, text modality, video modality, and audio modality. Denote taking the logarithmic transformation of , which is used to smooth the impact of the recommended content adaptability score. When is small, the score changes rapidly; when is large, the score change tends to be gentle, avoiding overemphasizing the modality with a high adaptability score. Denote inputting the adaptability factor \(\mathbf{c}\) of the user's current context into the sigmoid function to map its value to the range \((0, 1]\). When is large, the denominator approaches 1, enhancing the comprehensive score of this modality. When is small, the denominator increases, reducing the comprehensive score of this modality.

[0038] According to the selected optimal modality, convert the optimal recommended content of personalized information service into display forms of text modality, image modality, voice modality, and video modality, and dynamically adjust the push method for pushing according to the user's device type, network condition, and push timeliness. It should be noted that according to the adjustment of the device type, different device types determine the suitable modalities and push methods. For smartphones, the applicable modalities are text, pictures, short videos, and voice; for tablets / PCs, the applicable modalities are images, videos, and long texts; for smart speakers, the applicable modality is voice; for smart TVs, the applicable modality is video. According to the adjustment of the network condition, the network condition restricts the transmission efficiency of the recommended content, and dynamically adjusts the content modality and push method. For low-bandwidth networks, prefer text modality or compressed image modality to avoid voice and video modalities that consume high bandwidth, and push content links in a lightweight manner; for high-bandwidth networks, prefer to push video modality or high-resolution image modality and directly push the complete content. According to the adjustment of the push timeliness, the push timeliness determines the timeliness of the content and the best time for the user to receive it. For immediate push, it is for content that requires the user to respond quickly or has a short time limit, such as immediate notifications or message pushes; for delayed push, it is for non-urgent long-term content, such as scheduled pushes or batch pushes.

[0039] This embodiment also provides an information service system based on big data, including: a data collection module, a weighted multimodal feature generation module, a user portrait generation module, a recommended content module, and a dynamic push module; the data collection module is used to collect multimodal data; the weighted multimodal feature generation module is used to extract multimodal feature vectors and assign weights to the multimodal features based on the collected multimodal data to generate a weighted multimodal feature set; the user portrait generation module is used to perform feature fusion based on the weighted multimodal feature set to generate a user portrait vector; the recommended content module is used to generate the optimal recommended content of personalized information services by using a reinforcement learning model according to the user portrait vector; the dynamic push module is used to perform dynamic push based on the optimal recommended content of personalized information services.

[0040] This embodiment also provides a computer device applicable to the situation of the information service control method based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the information service control method based on big data proposed in the above embodiment.

[0041] This computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0042] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the information service control method based on big data proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0043] In summary, the present invention: collects multi-modal information such as text, images, voice, and environmental data, extracts multi-modal feature vectors and dynamically assigns weights to generate a weighted multi-modal feature set; then uses the Transformer model to deeply fuse the multi-modal features to generate a high-precision user portrait vector, so as to comprehensively and accurately characterize the user's interests, behavior preferences, and current environmental status; combines with a reinforcement learning model to dynamically generate the optimal recommended content of personalized information services based on the user portrait vector, and displays it through an optimized dynamic push method. Through the generation of weighted multi-modal features, the dynamic modeling of the non-linear relationship between modalities is realized, and the accuracy of feature fusion is enhanced; through the deep fusion of user portraits and the introduction of reinforcement learning algorithms, the problem that the recommended content in the prior art lacks real-time feedback optimization is solved, and finally the accuracy, adaptability, and intelligence level of information services are improved.

[0044] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An information service control method based on big data, characterized in that: including collecting multimodal data; extracting multimodal feature vectors based on the collected multimodal data and assigning weights to the multimodal features to generate a weighted multimodal feature set; performing feature fusion based on the weighted multimodal feature set to generate a user profile vector; using a reinforcement learning model to generate the optimal recommended content for personalized information services according to the user profile vector; performing dynamic push based on the optimal recommended content for personalized information services.

2. The information service control method based on big data according to claim 1, wherein: The multimodal data includes text data, image data, voice data, and environmental data.

3. The information service control method based on big data according to claim 2, wherein: The steps for extracting multimodal feature vectors based on the collected multimodal data are as follows performing word segmentation, removing stop words on the multimodal data, and extracting semantic features to generate text feature vectors; using a pre-trained convolutional neural network to extract image feature vectors; converting voice data to text and combining with a sentiment analysis model to extract user emotion features to generate voice feature vectors; performing data standardization processing on the environmental data and extracting environmental feature vectors.

4. The information service control method based on big data according to claim 3, wherein: The steps for assigning weights to the multimodal features to generate a weighted multimodal feature set are as follows dynamically calculating the weight of each modality based on the text feature vector, image feature vector, voice feature vector, and environmental feature vector, and the expression is ； Among them, represents the weight of the th modality, represents the historical interaction weight of the th modality, represents the natural exponential function, represents the index for text, image, speech, and environmental modalities; Apply the obtained modal weight vector to the corresponding modal feature vectors to generate a weighted multi-modal feature set, with the expression: ； Among them, represents the weighted multi-modal feature set, represents the feature vector of the 5. The information service control method based on big data according to claim 4, wherein: The steps for performing feature fusion based on the weighted multimodal feature set to generate a user profile vector are as follows using a Transformer model based on the weighted multimodal feature set to calculate the fusion feature representation of each modality, and the expression is ； Among them, represents the fused feature representation of the th modality, represents the feature similarity between the th modality and the th modality, represents the attention score between the th modality and the th modality, represents the attention score between the th modality and the th modality, represents the feature vector of the th modality; Fusion feature representation of all modalities Concatenate them, and generate a unified user portrait vector through a fully connected layer and a non-linear activation function. The expression is as follows: ； Among them, represents the user portrait vector, represents the activation function, represents the weight matrix of the fully connected layer, represents the bias term, represents the fused feature of the text modality, represents the fused feature of the image modality, represents the fused feature of the speech modality, represents the fused feature of the environmental modality.

6. The information service control method based on big data according to claim 5, wherein: The steps for using a reinforcement learning model to generate the optimal recommended content for personalized information services according to the user profile vector are as follows The generated user portrait vector is decomposed into a multi-dimensional feature subspace, which is divided into the static feature subspace of the user and the dynamic feature subspace ; Send and into the feature embedding layer respectively to generate the state of the reinforcement learning model. The expression is as follows: ； Among them, represents the state representation of reinforcement learning, represents the embedding matrix of static features, represents the embedding matrix of dynamic features; initializing the action value function using the DQN algorithm based on the state representation of reinforcement learning, and the optimization goal is to maximize the action value, and the expression is ； Among them, represents the current state and the action value of the execution action of represents the learning rate, represents the immediate reward value, represents the discount factor, represents the future state of the optimal action corresponding to the maximum action value, represents the current action, represents the future state, represents the next state and is one of all actions in the Select the action with the maximum action value according to the optimized action value function As the recommended content, the expression is: ； Among them, represents the maximum action, represents the current state and the action to be executed and the action value represents the candidate set of recommended content; The maximum action will be obtained , as the optimal recommended content for personalized information services.

7. The information service control method based on big data according to claim 6, characterized in that: The steps for performing dynamic push based on the optimal recommended content for personalized information services are as follows using an optimization model to select the optimal modality based on the optimal recommended content for personalized information services, and the expression is ； Among them, represents the selected optimal modality, represents the set of all available modalities, represents the preference weight for modality in the user profile, represents the adaptability score of the recommended content on modality and represents the adaptability factor of the user's current context to modality . According to the selected optimal modality, converting the optimal recommended content for personalized information services into display forms of text modality, image modality, voice modality, and video modality, and dynamically adjusting the push method for pushing according to the user's device type, network condition, and push timeliness.

8. An information service system based on big data, based on the information service control method based on big data according to any one of claims 1 to 7, characterized in that: including a data collection module, a weighted multimodal feature generation module, a user profile generation module, a recommended content module, and a dynamic push module; The data collection module is used to collect multimodal data; The weighted multimodal feature generation module is used to extract multimodal feature vectors based on the collected multimodal data and assign weights to the multimodal features to generate a weighted multimodal feature set; The user profile generation module is used to perform feature fusion based on the weighted multimodal feature set to generate a user profile vector; The recommended content module is used to use a reinforcement learning model to generate the optimal recommended content for personalized information services according to the user profile vector; The dynamic push module is used to perform dynamic push based on the optimal recommended content for personalized information services.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, the steps of the information service control method based on big data according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the information service control method based on big data according to any one of claims 1 to 7 are implemented.

Images