Content recommendation system, method, apparatus, electronic device, and storage medium

Abstract

The present disclosure provides a content recommendation system, method, device, electronic equipment and storage medium. The system comprises a network quality evaluation module, a user clustering module, a content preprocessing module and an abstract generation module. The network quality evaluation module is configured to determine the network quality level of a terminal based on the obtained network quality information of the terminal, and send the network quality level to the abstract generation module. The user clustering module is configured to determine the user category based on the obtained feature data of the user, analyze the feature data, and send the user category to the abstract generation module. The content preprocessing module is configured to generate a preprocessed abstract by differentiating the types of the content to be recommended, and send the preprocessed abstract to the abstract generation module. The abstract generation module is configured to receive the network quality level, the user category and the preprocessed abstract, process the preprocessed abstract according to the network quality level and the user category, and generate personalized recommended content.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and more particularly to a content recommendation system, method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the development of internet technology, information is growing explosively, and users face the problem of difficulty in selecting information during the process of obtaining information. To solve this problem, content recommendation systems have emerged. Their function is to filter information according to user needs, reduce the cost of information filtering for users, and have been widely used in various business scenarios for individual users, such as video, cloud storage, and communication.

[0003] However, current content recommendation systems determine recommended content based solely on user needs, and their processing methods for recommended content are simplistic, resulting in insufficient adaptation between recommended content and the user's actual reception scenario, and low user click-through rates. Summary of the Invention

[0004] This disclosure provides a content recommendation system, method, apparatus, electronic device, and storage medium. Its main objective is to address the problem that existing content recommendation systems cannot simultaneously consider both terminal network quality and the personalization of recommended content.

[0005] According to a first aspect of this disclosure, a content recommendation system is provided, the system comprising: a network quality assessment module, a user segmentation module, a content preprocessing module, and a summary generation module; wherein... The network quality assessment module is used to determine the network quality level of the terminal based on the obtained network quality information of the terminal, and send the network quality level to the summary generation module; The user segmentation module is used to analyze the acquired user feature data to determine the user category, and send the user category to the summary generation module. The content preprocessing module is used to perform differentiated processing based on the type of content to be recommended to generate a preprocessed summary, and send the preprocessed summary to the summary generation module; The summary generation module is used to receive the network quality level, the user category, and the preprocessed summary, and process the preprocessed summary according to the network quality level and the user category to generate personalized recommendation content.

[0006] In some embodiments, the network quality assessment module includes: an acquisition unit, a processing unit, and a determination unit; wherein, The acquisition unit is used to acquire the network transmission parameters and wireless signal parameters of the terminal; and send the network transmission parameters and wireless signal parameters to the processing unit. The processing unit is configured to process the network transmission parameters and the wireless signal parameters to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters; and send the standardized scores to the determining unit. The determining unit is used to calculate the standardized scores corresponding to the network transmission parameters and the wireless signal parameters based on a preset weight allocation rule, obtain a comprehensive score, and determine the network quality level of the terminal based on the comprehensive score.

[0007] In some embodiments, the content preprocessing module includes: a text processing unit, a matching unit, a scoring unit, and a summary generation unit; wherein, The text processing unit is used to extract text and split sentences into the content to be recommended, obtain a sentence list, and send the sentence list to the matching unit and / or the scoring unit. The matching unit is configured to receive the sentence list when the content to be recommended contains an image, determine the relevance parameters between each sentence in the sentence list and the image, and send the relevance parameters to the summary generation unit. The scoring unit is used to receive the sentence list, determine the text importance score of each sentence in the sentence list, and send the text importance score to the summary generation unit; The summary generation unit is used to select target sentences and generate the preprocessed summary based on the relevance parameter and the text importance score when the content to be recommended contains the image. The summary generation unit is further configured to generate the preprocessed summary by selecting target sentences based on the text importance score when the content to be recommended does not contain the image.

[0008] In some embodiments, the system further includes: an adjustment module; The adjustment module is used to construct a training dataset, train the large model called by the summary generation module to obtain the trained large model, and send the trained large model to the summary generation module.

[0009] In some embodiments, the summary generation module is further configured to: The trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0010] In some embodiments, the system further includes: an edge deployment module; The edge deployment module is used to call the large model of the edge nodes in the corresponding area based on the geographical location information of the terminal.

[0011] According to a second aspect of this disclosure, a content recommendation method is provided, the content recommendation method being applied to the content recommendation system described in the first aspect above, comprising: The network quality level of the terminal is determined based on the obtained network quality information of the terminal; Obtain user feature data, analyze the feature data, and determine the user category corresponding to the user; Based on the type of content to be recommended, a preprocessed summary is generated through differentiated processing. Based on the network quality level and the user category, the preprocessed summary is processed to generate personalized recommendation content.

[0012] In some embodiments, determining the network quality level of the terminal based on the acquired network quality information of the terminal includes: Obtain the network transmission parameters and wireless signal parameters of the terminal; The network transmission parameters and the wireless signal parameters are processed to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters; The standardized scores are calculated based on preset weighting rules to obtain a comprehensive score; The network quality level of the terminal is determined based on the comprehensive score.

[0013] In some embodiments, the step of generating a preprocessed summary by performing differentiated processing based on the type of content to be recommended includes: The text to be recommended is extracted and the sentences are split to obtain a list of sentences; Determine the text importance score for each sentence in the sentence list; Determine whether the content to be recommended contains images; If the content to be recommended contains the image, then the relevance parameters between each sentence in the sentence list and the image are determined, and the target sentence is selected based on the relevance parameters and the text importance score to generate the preprocessed summary; If the content to be recommended does not contain the image, then the target sentence is selected based on the text importance score to generate the preprocessed summary.

[0014] In some embodiments, processing the preprocessed summary based on the network quality level and the user category to generate personalized recommendation content includes: Based on the network quality level and the user category, a large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0015] In some embodiments, after processing the preprocessed summary using a large model based on the network quality level and the user category to generate the personalized recommendation content, the method further includes: A training dataset is constructed, and the large model is trained to obtain the trained large model.

[0016] In some embodiments, the step of processing the preprocessed summary using a large model based on the network quality level and the user category to generate the personalized recommendation content includes: Based on the network quality level and the user category, the trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0017] In some embodiments, the step of processing the preprocessed summary using a large model based on the network quality level and the user category to generate the personalized recommendation content includes: Obtain the geographical location information of the terminal; The large model of the corresponding regional edge node is invoked based on the geographical location information.

[0018] In some embodiments, obtaining user feature data, analyzing the feature data, and determining the user category corresponding to the user includes: Obtain the user's feature data; The feature data is cleaned to remove outliers and duplicate data, resulting in cleaned feature data. The cleaned feature data is processed to obtain standardized feature data; Based on preset feature selection rules, key features are extracted from the standardized feature data; Analyze the key features to obtain at least one clustering result; Based on the at least one clustering result, determine the user category corresponding to the user.

[0019] According to a third aspect of this disclosure, a content recommendation device is provided, the content recommendation device being configured in the content recommendation system described in the first aspect above, comprising: The first determining unit is used to determine the network quality level of the terminal based on the network quality information of the acquired terminal. The second determining unit is used to acquire the user's feature data, analyze the feature data, and determine the user category corresponding to the user. The first generation unit is used to generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended. The second generation unit is used to process the preprocessed summary based on the network quality level and the user category to generate personalized recommendation content.

[0020] In some embodiments, the first determining unit includes: The first acquisition module is used to acquire the network transmission parameters and wireless signal parameters of the terminal. The first determining module is used to process the network transmission parameters and the wireless signal parameters to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters; The second determining module is used to calculate the standardized score based on a preset weight allocation rule to obtain a comprehensive score; The third determining module is used to determine the network quality level of the terminal based on the comprehensive score.

[0021] In some embodiments, the first generating unit includes: The fourth determining module is used to extract text and split sentences from the content to be recommended to obtain a list of sentences; The fifth determining module is used to determine the text importance score of each sentence in the sentence list; The judgment module is used to determine whether the content to be recommended contains images; The sixth determining module is used to determine the relevance parameters between each sentence in the sentence list and the image when the content to be recommended contains the image, select the target sentence based on the relevance parameters and the text importance score, and generate the preprocessed summary; The seventh determining module is used to select a target sentence based on the text importance score and generate the preprocessed summary when the content to be recommended does not contain the image.

[0022] In some embodiments, the second generating unit includes: The generation module is used to process the preprocessed summary by calling a large model based on the network quality level and the user category, and generate the personalized recommendation content.

[0023] In some embodiments, the apparatus further includes: The eighth determining module is used to construct a training dataset and train the large model after the generation module processes the preprocessed summary based on the network quality level and the user category, thereby obtaining the trained large model.

[0024] In some embodiments, the generation module is further configured to: Based on the network quality level and the user category, the trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0025] In some embodiments, the generation module includes: The acquisition submodule is used to acquire the geographical location information of the terminal; The submodule is invoked to invoke the large model of the corresponding regional edge node based on the geographic location information.

[0026] In some embodiments, the second determining unit includes: The second acquisition module is used to acquire the user's feature data; The ninth determining module is used to clean the feature data, remove outliers and duplicate data, and obtain cleaned feature data. The tenth determining module is used to process the cleaned feature data to obtain standardized feature data; The extraction module is used to extract key features from the standardized feature data based on preset feature filtering rules; The eleventh determination module is used to analyze the key features and obtain at least one clustering result; The twelfth determining module is used to determine the user category corresponding to the user based on the at least one clustering result.

[0027] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the second aspect above.

[0028] According to a fifth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the second aspect above.

[0029] According to a sixth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the second aspect above.

[0030] In summary, compared with related technologies, the solution disclosed herein can obtain the network quality information of the terminal and determine the network quality level through the network quality assessment module, obtain user feature data and determine user categories through the user segmentation module, generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended through the content preprocessing module, and then receive the network quality level, user category and preprocessed summary through the summary generation module, process the preprocessed summary in combination with the network quality level and user category to generate personalized recommendation content. This ensures that the generated personalized recommendation content is adapted to the network conditions of the terminal and meets the needs of users, thereby improving the adaptability of personalized recommendation content.

[0031] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0032] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram of the structure of a content recommendation system provided in an embodiment of the present disclosure; Figure 2 This is a schematic diagram of another content recommendation system provided in an embodiment of the present disclosure; Figure 3 A schematic flowchart illustrating a content recommendation method provided in an embodiment of this disclosure; Figure 4 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 5 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 6 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 7 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 8 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 9 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 10 A flowchart illustrating another content recommendation method provided in this embodiment of the disclosure; Figure 11 This is a schematic diagram of the structure of a content recommendation device provided in an embodiment of the present disclosure; Figure 12This is a schematic diagram of another content recommendation device provided in an embodiment of the present disclosure; Figure 13 This is a schematic block diagram of an example electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0033] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0034] The following description, with reference to the accompanying drawings, describes a content recommendation system, method, apparatus, electronic device, and storage medium according to embodiments of the present disclosure.

[0035] Figure 1 This is a schematic diagram of the structure of a content recommendation system provided in an embodiment of this disclosure.

[0036] like Figure 1 As shown, the system includes: The module comprises: Network Quality Assessment Module 1, User Segmentation Module 2, Content Preprocessing Module 3, and Summary Generation Module 4; among which, The network quality assessment module 1 is used to determine the network quality level of the terminal based on the obtained network quality information of the terminal, and send the network quality level to the summary generation module 4.

[0037] In some embodiments, the network quality information of the terminal includes network transmission-related data, signal reception-related data, etc. generated by the terminal during wireless communication. The network quality assessment module 1 receives the network quality information reported by the terminal, analyzes the network quality information using a preset assessment logic, and determines the network quality level of the terminal. The network quality level can be divided into multiple levels. After determining the network quality level, the network quality assessment module 1 sends the network quality level to the summary generation module 4 through the internal information transmission channel of the system to ensure that the summary generation module 4 can obtain the current network quality level information of the terminal.

[0038] The user segmentation module 2 is used to analyze the acquired user feature data to determine the user category, and send the user category to the summary generation module 4.

[0039] In some embodiments, user characteristic data includes, but is not limited to, user basic attribute data, behavioral data, and preference data. Basic attribute data may include information such as the user's age and gender; behavioral data may include information such as the user's content browsing history, content click records, and content dwell time; preference data may include information such as the content areas the user has long followed and the types of content they frequently interact with. After acquiring the aforementioned characteristic data through the system's data acquisition interface, the user segmentation module 2 processes the characteristic data using preset analysis methods. These analysis methods include, but are not limited to, classification methods based on feature similarity and categorization methods based on user behavior patterns. These analysis methods group users with similar characteristics into the same category, determining the user category corresponding to each user. After determining the user category, the user segmentation module 2 sends the user category to the summary generation module 4 through the system's internal information transmission channel.

[0040] The content preprocessing module 3 is used to perform differentiated processing based on the type of content to be recommended to generate a preprocessed summary, and send the preprocessed summary to the summary generation module 4.

[0041] In some embodiments, the types of content to be recommended include content containing images and content consisting solely of plain text. Content containing images refers to content that includes both text and image information, while content consisting solely of text refers to content that contains only text information without any image information. The content preprocessing module 3 identifies the type of content to be recommended and applies different processing methods based on the identified type, i.e., it performs differentiated processing: for content containing images, it extracts information by combining the relationship between image and text information in the content, the logical structure of the text information itself, and key information; for content consisting solely of plain text, it extracts information based on the logical structure of the text information itself and key information, generating a preprocessed summary that reflects the key information of the content to be recommended. After generating the preprocessed summary, the content preprocessing module 3 sends the preprocessed summary to the summary generation module 4 through the internal information transmission channel of the system.

[0042] The summary generation module 4 is used to receive the network quality level, the user category and the preprocessed summary, and process the preprocessed summary according to the network quality level and the user category to generate personalized recommendation content.

[0043] In some embodiments, the summary generation module 4 receives the network quality level sent by the network quality assessment module 1, the user category sent by the user segmentation module 2, and the preprocessed summary sent by the content preprocessing module 3. After receiving the above information, the summary generation module 4 determines the adjustment dimensions of the preprocessed summary based on the network quality level: on the one hand, the network quality level reflects the current network transmission capability of the terminal, and different levels correspond to different transmission adaptation requirements, thereby determining the adjustment direction of the preprocessed summary in terms of information density and data volume; on the other hand, images increase the overall data volume of personalized recommendation content, and the final presentation form of personalized recommendation content is determined according to the network quality level. Under the network quality level corresponding to high network transmission capability, the terminal can receive and... Personalized recommendations containing images can be loaded, and images can be added to the personalized recommendations. However, under low network quality levels (corresponding to low network transmission capacity), images are not added to avoid transmission delays or loading failures. The summary generation module 4 clarifies the style adaptation dimension of the preprocessed summary based on user categories. User categories reflect user characteristics and preferences, with different categories corresponding to different content expression preferences. Based on this, the adaptation direction of the preprocessed summary in terms of language style and content focus is determined. The adjustment direction corresponding to the network quality level, the image addition judgment, and the style adaptation direction corresponding to the user category are combined to integrate the preprocessed summary, ultimately forming personalized recommendations that meet both terminal network transmission conditions and user preferences.

[0044] For example, if the network quality level corresponds to a lower network transmission capacity, the adjustment direction is to reduce information density to decrease data volume, and no images are added; if the user category is teenagers, the style adaptation direction is to adopt a lively and popular expression style; combining both, the preprocessed summary is simplified into content mainly composed of short sentences, using the expression style commonly used by teenagers, and no images are added, generating the final personalized recommendation content. If the network quality level corresponds to a higher network transmission capacity, the adjustment direction is to retain a higher information density, and images can be added; combined with the style adaptation direction for teenagers, the expression style commonly used by teenagers is used, and images are added to generate the final personalized recommendation content.

[0045] In summary, compared with related technologies, the solution disclosed herein can obtain the network quality information of the terminal and determine the network quality level through the network quality assessment module, obtain user feature data and determine user categories through the user segmentation module, generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended through the content preprocessing module, and then receive the network quality level, user category and preprocessed summary through the summary generation module, process the preprocessed summary in combination with the network quality level and user category to generate personalized recommendation content. This ensures that the generated personalized recommendation content is adapted to the network conditions of the terminal and meets the needs of users, thereby improving the adaptability of personalized recommendation content.

[0046] Figure 2 A schematic diagram of the structure of a content recommendation system provided in an embodiment of this disclosure is further shown.

[0047] like Figure 2 As shown, the network quality assessment module 1 includes: an acquisition unit 11, a processing unit 12, and a determination unit 13; wherein, The acquisition unit 11 is used to acquire the network transmission parameters and wireless signal parameters of the terminal; and send the network transmission parameters and wireless signal parameters to the processing unit 12.

[0048] In some embodiments, network transmission parameters are parameters reflecting the data transmission efficiency between the terminal and the network, including but not limited to bandwidth, latency, and packet loss rate; wireless signal parameters are parameters reflecting the strength and quality of the wireless signal received by the terminal, including but not limited to Reference Signal Receiving Power (RSRP) and Reference Signal Receiving Quality (RSRQ). The acquisition unit 11 acquires the above parameters in two ways, including but not limited to: first, by calling the terminal's open system interface, such as the TELEPHONY_SERVICE interface of the Android system or the CoreTelephony framework of the iOS system, to read the network transmission parameters and wireless signal parameters collected by the terminal; second, by receiving parameter data periodically reported by the terminal, with the terminal collecting parameters at preset time intervals (e.g., 10 seconds / time) and sending them to the acquisition unit 11. After collecting the network transmission parameters and wireless signal parameters, the acquisition unit 11 sends them to the processing unit 12.

[0049] The processing unit 12 is used to process the network transmission parameters and the wireless signal parameters to obtain the standardized scores corresponding to the network transmission parameters and the wireless signal parameters; and send the standardized scores to the determining unit 13.

[0050] In some embodiments, the processing unit 12 processes network transmission parameters and wireless signal parameters, converting parameters with different dimensions and numerical ranges into a unified range (e.g., 0-100 points) for easier calculation of the overall score. The standardization process employs methods including, but not limited to, the min-max standardization method. Specifically, for each parameter, the optimal and worst values ​​are preset (e.g., optimal bandwidth is 100Mbps, worst is 1Mbps; optimal latency is 10ms, worst is 200ms). Then, the standardized score for the parameter is calculated using the formula (actual parameter value - worst parameter value) / (optimal parameter value - worst parameter value) × 100. Higher values ​​for bandwidth, RSRP, and RSRQ correspond to higher scores, while lower values ​​for latency and packet loss rate correspond to higher scores. After completing the standardization process for all parameters, the processing unit 12 compiles the standardized scores for each parameter into a data list and sends it to the determining unit 13.

[0051] The determining unit 13 is used to calculate the standardized scores corresponding to the network transmission parameters and the wireless signal parameters based on a preset weight allocation rule, obtain a comprehensive score, and determine the network quality level of the terminal based on the comprehensive score.

[0052] In some embodiments, the preset weight allocation rule refers to setting different weight percentages for the standardized scores of each parameter. The weight percentages are determined according to the degree of influence of the parameter on network quality. For example, bandwidth weight 0.3, latency weight 0.2, packet loss rate weight 0.2, RSRP weight 0.15, and RSRQ weight 0.15. The above is only an illustrative example and is not a limitation on specific parameters and weights.

[0053] The method for calculating the comprehensive score in Unit 13 is the sum of the standardized scores of each parameter × their corresponding weights, i.e., comprehensive score = (bandwidth score × 0.3) + (latency score × 0.2) + (packet loss rate score × 0.2) + (RSRP score × 0.15) + (RSRQ score × 0.15).

[0054] Network quality levels are divided into multiple tiers based on a comprehensive score. For example, a comprehensive score ≥ 80 points is tier 1 (excellent network quality), 60 points ≤ comprehensive score < 80 points is tier 2 (good network quality), and a comprehensive score < 60 points is tier 3 (average network quality). The determination unit 13 calculates the comprehensive score to determine the terminal's network quality level. The above is merely an illustrative example and not a limitation on the specific tiers of network quality levels.

[0055] Through the above structure, the acquisition unit, processing unit, and determination unit work together to determine the network quality level of the terminal, providing a reliable network condition basis for the subsequent preprocessing of the summary by the summary generation module.

[0056] like Figure 2 As shown, the content preprocessing module 3 includes: a text processing unit 31, a matching unit 32, a scoring unit 33, and a summary generation unit 34; wherein, The text processing unit 31 is used to extract text and split sentences from the content to be recommended, obtain a list of sentences, and send the list of sentences to the matching unit 32 and / or the scoring unit 33.

[0057] In some embodiments, if the content to be recommended contains images, the text processing unit 31 extracts the text portion of the content using technologies including but not limited to optical character recognition (OCR). If the content is plain text, all text is extracted directly. The text processing unit 31 performs sentence splitting on the text in the content to be recommended. Based on preset punctuation rules (such as using periods, question marks, and exclamation marks as splitting markers, excluding commas, semicolons, and other non-sentence-ending punctuation), the extracted text is split into independent sentences, forming a sentence list. The sending logic of the sentence list is determined according to the type of content to be recommended: if the content to be recommended contains images, the text processing unit 31 sends the sentence list to both the matching unit 32 and the scoring unit 33; if the content is plain text, only the sentence list is sent to the scoring unit 33.

[0058] The matching unit 32 is configured to receive the sentence list when the content to be recommended contains an image, determine the relevance parameters between each sentence in the sentence list and the image, and send the relevance parameters to the summary generation unit 34.

[0059] In some embodiments, the relevance parameter refers to the degree of semantic similarity between the sentence and the image, and its value range is typically 0-1 (0 indicates no relevance, 1 indicates high relevance). Matching unit 32 calculates this parameter using a cross-modal matching model. The model is trained through contrastive learning, mapping the visual features of the image (such as color, outline, and scene) and the textual features of the sentence (such as keywords and semantic vectors) to the same shared embedding space, and then using the cosine similarity formula... The similarity between the image embedding vector (B) and the embedding vectors of each sentence (A) is calculated, and this similarity is the relevance parameter. After the matching unit 32 calculates the relevance parameters of all sentences, it sends them to the summarization generation unit 34.

[0060] The scoring unit 33 is used to receive the sentence list, determine the text importance score of each sentence in the sentence list, and send the text importance score to the summary generation unit 34.

[0061] In some embodiments, the text importance score reflects the criticality of a sentence within the text of the content to be recommended; a higher score indicates that the sentence contains more key information. The scoring unit 33 calculates the score by constructing a sentence association graph: taking each sentence in the sentence list as a node, it calculates the semantic similarity (e.g., cosine similarity based on TF-IDF vectors) between any two nodes (sentences), uses this similarity as the weight of the edge between nodes, and forms a sentence association graph; it then uses a centrality algorithm (including but not limited to the PageRank algorithm) to traverse the association graph. Calculate the centrality score of each node, which is the text importance score of the sentence (range 0-100), where p(u) is the centrality score of node u, p(v) is the centrality score of node v, adj[u] is the set of nodes adjacent to u, and deg(v) is the degree of node v.

[0062] After scoring unit 33 completes the text importance scoring of all sentences, it sends the results to summary generation unit 34.

[0063] The summary generation unit 34 is used to select target sentences and generate the preprocessed summary based on the relevance parameters and the text importance score when the content to be recommended contains the image.

[0064] In some embodiments, the summary generation unit 34 receives the relevance parameters sent by the matching unit 32 and the text importance scores sent by the scoring unit 33, and calculates the comprehensive score of each sentence according to a preset weighting rule (e.g., relevance parameter weight 0.4, text importance score weight 0.6), the formula being scores=0.6×p(u)+0.4×similarity(A,B); the sentences are sorted from high to low according to the comprehensive scores, and the top N sentences are selected as target sentences (N is preset according to the length of the text to be recommended, e.g., the first 2 sentences are selected if the text length is less than 500 words, and the first 3 sentences are selected if the text length is more than 500 words); the target sentences are concatenated in the order in the text to be recommended to form a preprocessed summary. The above is only an illustrative example and is not a limitation on specific weights.

[0065] The summary generation unit 34 is further configured to generate the preprocessed summary by selecting target sentences based on the text importance score when the content to be recommended does not contain the image.

[0066] In some embodiments, the summary generation unit 34 receives the text importance score sent by the scoring unit 33, sorts the sentences from high to low according to the score, selects the top N sentences (N is preset according to the length of the text to be recommended, such as the first 2 sentences for texts with a length of less than 500 words, and the first 3 sentences for texts with a length of more than 500 words) as target sentences; and splices the target sentences in the order in the text to be recommended to form a preprocessed summary.

[0067] The above structure enables differentiated processing of recommended content, ensuring that the preprocessed summary not only matches the key information of the content but also adapts to the content type.

[0068] like Figure 2 As shown, the system also includes: adjustment module 5; The adjustment module 5 is used to construct a training dataset, train the large model called by the summary generation module to obtain the trained large model, and send the trained large model to the summary generation module 4.

[0069] In some embodiments, the training dataset constructed by the adjustment module 5 includes preprocessed summaries, corresponding user categories, and personalized recommendation content samples adapted to user categories, which are formed by collecting historical valid data and standardizing them; during training, a parameter-efficient fine-tuning method (including but not limited to LoRA) is used to optimize the parameters of the large model with the preprocessed summaries and user categories as inputs and the adapted recommendation content as outputs. After the large model converges, the trained large model is sent to the summary generation module 4.

[0070] The above structure enables the trained large model to adapt to the preferences of different user categories, improving the accuracy of personalized recommendation content generated by the summary generation module.

[0071] In a specific embodiment of this disclosure, the summary generation module 4 is further configured to: The trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0072] In some embodiments, the summary generation module 4 matches the trained large model sent by the adjustment module 5 with the user categories sent by the user segmentation module 2; the preprocessed summary and network quality level are integrated into the model input data according to a preset format (e.g., "preprocessed summary: [content]; network quality level: [level]"); the trained large model generates personalized recommendation content based on the input data, while adapting to the user category style (e.g., business users correspond to concise and professional expressions) and adjusting the content data volume in combination with the network quality level (e.g., low network quality level corresponds to concise text and does not contain images).

[0073] Through the above structure, the summary generation module calls the trained large model, which can further improve the accuracy and practicality of personalized recommendation content.

[0074] like Figure 2 As shown, the system also includes: an edge deployment module 6; The edge deployment module 6 is used to call the large model of the edge nodes in the corresponding area based on the geographical location information of the terminal.

[0075] In some embodiments, the terminal's geographic location information is obtained through the terminal's built-in positioning function, including but not limited to GPS positioning and base station positioning. Edge nodes are divided according to geographical regions, such as provincial, municipal, or district-level administrative units. Each edge node deploys at least one large model. After receiving the terminal's geographic location information, the edge deployment module 6 determines the region to which the terminal belongs by parsing the geographic location information and finds the corresponding edge node. Through a preset low-latency communication interface (such as a 5G edge computing interface or a dedicated local area network interface), it sends a call request to the edge node. The request contains data such as a pre-processed summary, network quality level, and user category. The large model of the edge node processes the data and returns personalized recommendation content.

[0076] The above structure enables large model calls to be completed locally, reducing the time consumption of data transmission across regions and lowering the latency of personalized recommendation content generation.

[0077] Figure 3 This is a flowchart illustrating a content recommendation method provided in an embodiment of the present disclosure, wherein the content recommendation method is applied to the aforementioned content recommendation system.

[0078] like Figure 3 As shown, the method includes steps 101-104.

[0079] Step 101: Determine the network quality level of the terminal based on the obtained network quality information of the terminal.

[0080] In some embodiments, the network quality information of a terminal refers to data reflecting the network status generated by the terminal during its interaction with the network, including but not limited to network transmission stability data and signal reception strength data. Methods for obtaining the terminal's network quality information include receiving information actively reported by the terminal or reading the terminal's network status data in real time through a system interface. When determining the network quality level based on the network quality information, a preset evaluation standard must be adopted. Multiple level ranges should be divided according to key indicators of the network quality information (such as data transmission rate and signal stability). For example, network quality can be divided into three levels: "high," "medium," and "low," with each level corresponding to a different range of indicators. The above is merely an illustrative example and not a limitation on the specific method of dividing network quality levels.

[0081] By comparing the acquired network quality information with a preset range of levels, the network quality level of the terminal can be determined. For example, when the terminal's network quality information shows that the data transmission rate is consistently higher than 50Mbps and the signal is stable, its network quality level is determined to be "high"; when the terminal's network quality information shows that the data transmission rate is lower than 10Mbps and the signal fluctuates significantly, its network quality level is determined to be "low". The above is only an illustrative example and is not a limitation on the specific method of determining network quality levels.

[0082] Step 102: Obtain user feature data, analyze the feature data, and determine the user category corresponding to the user.

[0083] In some embodiments, user characteristic data refers to a data set that reflects user attributes, behavioral habits, and content preferences, including but not limited to user registration information (such as age range, occupation type), historical operation records (such as browsing content type, interaction frequency), and long-term areas of interest (such as technology, entertainment, education). When analyzing characteristic data, a pre-defined classification logic is used to divide users into different groups based on key dimensions in the characteristic data (such as content preference tendencies and interaction behavior patterns), with each group corresponding to a user category. The analysis process can be achieved by statistically analyzing the distribution of each dimension in the characteristic data or by using rule-based classification methods to group users with similar characteristics into the same category. For example, users who frequently browse educational content and interact primarily with accounts in the education field are classified as "education-interested users"; users who frequently view sports-related content are classified as "sports-interested users."

[0084] Step 103: Generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended.

[0085] In some embodiments, the types of content to be recommended include two: one is content containing images, i.e., content containing both text and image information; the other is pure text content, i.e., content containing only text information without image information. Differentiated processing is applied based on the type of content to be recommended, meaning different information extraction methods are used for the two types: for content containing images, the processing needs to consider the relationship between text and image information, extracting information that reflects the themes of both text and images; for pure text content, the processing is based solely on the logical structure and key content of the text itself. Through this differentiated processing, a preprocessed summary that summarizes the key information of the content to be recommended is generated. For example, for travel guide content containing images of "landscape scenery," the preprocessed summary will simultaneously cover the images in the content and the "best season to visit" and "landscape features" related to the images in the text; for pure text news reports, the preprocessed summary focuses on key textual information such as "event time," "key figures," and "main results."

[0086] Step 104: Based on the network quality level and the user category, process the preprocessed summary to generate personalized recommendation content.

[0087] In some embodiments, preprocessing summaries are processed based on network quality levels, primarily through adjustments to the content presentation. Higher network quality allows for the retention of more detailed information and longer content, including images. Lower network quality necessitates simplified information presentation, shorter content length, and removal of images from the preprocessing summary to adapt to network transmission capabilities. Processing preprocessing summaries based on user categories primarily involves adjusting content style and emphasis. Different user categories correspond to different content preferences, and the processing must ensure the content style matches the user category. Combining these two approaches integrates and adjusts the preprocessing summaries to ultimately create personalized recommendation content.

[0088] In summary, the content recommendation method provided in this disclosure can determine the network quality level based on terminal network quality information, determine the user category based on user feature data, generate a preprocessed summary according to the type of content to be recommended, and then combine the network quality level and user category to process the preprocessed summary to generate personalized recommendation content. This makes the generated personalized recommendation content both adaptable to terminal network conditions and meets user needs, thereby improving the adaptability of personalized recommendation content.

[0089] Figure 4 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 4 As shown, the content recommendation method includes steps 201-204.

[0090] Step 201: Obtain the network transmission parameters and wireless signal parameters of the terminal.

[0091] In some embodiments, network transmission parameters are parameters reflecting the data transmission efficiency between the terminal and the network, including but not limited to bandwidth, latency, and packet loss rate; wireless signal parameters are parameters reflecting the strength and quality of the wireless signal received by the terminal, including but not limited to RSRP and RSRQ. The methods for obtaining these parameters include, but are not limited to, the following two: one is to call the terminal's system interface, such as the TELEPHONY_SERVICE interface in Android or the CoreTelephony framework in iOS, to read the parameter data collected by the terminal; the other is to receive parameter data actively reported by the terminal at a preset period (e.g., once every 10 seconds). For example, the network transmission parameters obtained from the terminal are "bandwidth 30Mbps, latency 25ms, packet loss rate 0.2%", and the wireless signal parameters are "RSRP-80dBm, RSRQ-7dB".

[0092] Step 202: Process the network transmission parameters and the wireless signal parameters to obtain the standardized scores corresponding to the network transmission parameters and the wireless signal parameters.

[0093] In some embodiments, parameter processing involves converting parameters with different dimensions and numerical ranges into a unified score of 0-100. Processing methods include, but are not limited to, the min-max standardization method. Specifically, for each parameter, an optimal and worst-case value are preset (e.g., optimal bandwidth 100Mbps, worst-case 1Mbps; optimal latency 10ms, worst-case 200ms). For parameters such as bandwidth, RSRP, and RSRQ, where higher values ​​indicate better quality, the standardization score is calculated as "(actual parameter value - worst-case value) / (optimal value - worst-case value) × 100". For parameters such as latency and packet loss rate, where lower values ​​indicate better quality, the standardization score is calculated as "(worst-case value - actual parameter value) / (worst-case value - optimal value) × 100". For example, the standardized score for a bandwidth of 30Mbps is (30-1) / (100-1)×100≈29.3 points, and the standardized score for a latency of 25ms is (200-25) / (200-10)×100≈92.1 points. The above are only illustrative examples and are not a limitation on the specific parameter processing method.

[0094] Step 203: Calculate the standardized score based on the preset weight allocation rules to obtain the comprehensive score.

[0095] In some embodiments, the preset weight allocation rule refers to assigning different weight percentages to the standardized scores of each parameter, with the weights determined based on the degree of influence of each parameter on network quality. The overall score is calculated as the sum of the products of the standardized scores of each parameter and their corresponding weights.

[0096] Step 204: Determine the network quality level of the terminal based on the comprehensive score.

[0097] In some embodiments, network quality levels are divided based on a range of comprehensive scores. For example, a comprehensive score ≥ 80 points is Level 1 (Excellent network quality), 60 points ≤ comprehensive score < 80 points is Level 2 (Good network quality), and a comprehensive score < 60 points is Level 3 (Average network quality). When determining the level, the calculated comprehensive score is compared with a preset range; if it falls within the corresponding range, it is determined to be the corresponding level. The above is merely an illustrative example and not a limitation on the specific hierarchical division of network quality levels.

[0098] By employing the methods described above, the determination of network quality levels becomes objective and accurate, providing a reliable basis for network conditions in the generation of subsequent personalized recommendation content.

[0099] Figure 5 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 5 As shown, the content recommendation method includes steps 301-305.

[0100] Step 301: Extract text and split sentences from the content to be recommended to obtain a list of sentences.

[0101] In some embodiments, text extraction refers to obtaining all text information from the content to be recommended. If the content to be recommended contains images, the text content (such as titles and descriptive text within images) needs to be extracted using technologies including but not limited to optical character recognition (OCR), and combined with the plain text portion of the content to form a complete text. If the content to be recommended is plain text, then all the text of the content to be recommended is directly obtained to form a complete text. Sentence splitting refers to splitting the extracted complete text into independent sentences according to preset rules. The splitting rules are identified by sentence-ending punctuation marks (such as periods, question marks, and exclamation marks), ignoring commas, semicolons, and other non-sentence-ending punctuation. The ordered set of sentences formed after splitting is the sentence list.

[0102] Step 302: Determine the text importance score of each sentence in the sentence list.

[0103] In some embodiments, text importance scores are used to measure the criticality of a sentence within the content to be recommended; a higher score indicates that the information contained in the sentence is more important. Methods for determining text importance scores include, but are not limited to, constructing a sentence association graph: using each sentence in the sentence list as a node, calculating the semantic similarity (e.g., cosine similarity based on word vectors) between any two nodes, and using the similarity as the weight of the edge between nodes; traversing the association graph using a centrality algorithm (including but not limited to the PageRank algorithm), and... Calculate the centrality score for each node, which is the text importance score (range 0-100).

[0104] Step 303: Determine whether the content to be recommended contains images.

[0105] In some embodiments, the determination method is based on the metadata information or file structure of the content to be recommended: if the metadata of the content to be recommended contains an image format identifier (such as .jpg, .png), or the file structure of the content contains an independent image data block, it is determined that the content contains an image; if the metadata does not contain an image identifier and the file structure only contains text data, it is determined that the content does not contain an image.

[0106] If the content to be recommended contains the image, then proceed to step 304; if the content to be recommended does not contain the image, then proceed to step 305.

[0107] Step 304: Determine the relevance parameters between each sentence in the sentence list and the image, select the target sentence based on the relevance parameters and the text importance score, and generate the preprocessed summary.

[0108] In some embodiments, the relevance parameter reflects the semantic association between the sentence and the image, with a value ranging from 0 to 1 (0 indicating no association, 1 indicating high association). Methods for determining the relevance parameter include, but are not limited to, using a cross-modal matching model to map the visual features of the image (such as color distribution and object contours) and the textual features of the sentence (such as keyword vectors) to the same feature space, and then using the cosine similarity formula. The similarity between the image feature vector and the feature vectors of each sentence is calculated; this similarity is the relevance parameter. When selecting target sentences, the comprehensive score of each sentence is first calculated according to preset weights (e.g., relevance parameter weight 0.4, text importance score weight 0.6). Then, the sentences are sorted from high to low according to their comprehensive scores, and the top N sentences are selected (N is preset according to the text length of the content to be recommended). These sentences are then concatenated according to the order of the text in the content to be recommended to form a preprocessed summary.

[0109] Step 305: Select target sentences based on the text importance score and generate the preprocessed summary.

[0110] In some embodiments, when selecting target sentences, the sentence list is directly sorted from high to low according to the text importance score, and the top N sentences are selected (the preset rule for N is the same as in step 304). The sentences are then concatenated in the order of the text in the content to be recommended to form a preprocessed summary.

[0111] By using the above methods, we can ensure that the preprocessed summary not only reflects the key information of the content, but also adapts to the type of content to be recommended, providing accurate basic materials for the subsequent generation of personalized recommendation content.

[0112] Figure 6 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 6 As shown, the content recommendation method includes steps 401-404.

[0113] Step 401: Determine the network quality level of the terminal based on the obtained network quality information of the terminal.

[0114] Step 402: Obtain the user's feature data, analyze the feature data, and determine the user category corresponding to the user.

[0115] Step 403: Generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended.

[0116] For explanations of steps 401-403, please refer to [link / reference needed]. Figure 3 The detailed description of the embodiments disclosed herein will not be repeated here.

[0117] Step 404: Based on the network quality level and the user category, the preprocessed summary is processed using a large model to generate the personalized recommendation content.

[0118] In some embodiments, a large model refers to a pre-trained language model with text understanding and generation capabilities, including but not limited to generative models based on the Transformer architecture. Before calling the large model, the network quality level, user category, and preprocessed summary need to be integrated into the model input data according to a preset format. An example of the input format is "Network Quality Level: [Level Description]; User Category: [Category Name]; Preprocessed Summary: [Summary Content]". After receiving the input data, the large model parses the network quality level to determine the conciseness of the personalized recommendation content generation; it parses the user category to adjust the content style, adopting an appropriate language style for different user categories. Through the above processing, the large model reconstructs and optimizes the preprocessed summary to generate personalized recommendation content that conforms to network conditions and user preferences.

[0119] Figure 7 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 7 As shown, the recommended method for this content includes steps 501-505.

[0120] Step 501: Determine the network quality level of the terminal based on the obtained network quality information of the terminal.

[0121] Step 502: Obtain user feature data, analyze the feature data, and determine the user category corresponding to the user.

[0122] Step 503: Generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended.

[0123] Step 504: Based on the network quality level and the user category, call the large model to process the preprocessed summary and generate the personalized recommendation content.

[0124] For explanations of steps 501-504, please refer to [link / reference needed]. Figure 3 and Figure 6 The detailed description of the embodiments disclosed herein will not be repeated here.

[0125] Step 505: Construct a training dataset and train the large model to obtain the trained large model.

[0126] In some embodiments, each sample in the training dataset contains a preprocessed summary, the corresponding user category, and a personalized recommendation content sample adapted to the user category. This is achieved by collecting and cleaning historical valid data (removing mismatched or incorrectly labeled data). During training, an efficient parameter fine-tuning method (including but not limited to LoRA) is employed. Using "preprocessed summary + user category" as input and the adapted personalized recommendation content as output, the parameters are iteratively optimized based on a pre-trained language model (such as BART or GPT series) through a cross-entropy loss function and the Adam optimizer until the model converges, resulting in a large trained model.

[0127] By using the methods described above, large models can be tailored to the preferences of different user categories, thereby improving the accuracy of subsequent personalized recommendation content generation.

[0128] Figure 8 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further illustrated. For example... Figure 8 As shown, the recommended content method includes steps 601-604.

[0129] Step 601: Determine the network quality level of the terminal based on the obtained network quality information of the terminal.

[0130] Step 602: Obtain user feature data, analyze the feature data, and determine the user category corresponding to the user.

[0131] Step 603: Generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended.

[0132] For explanations of steps 601-603, please refer to [link / reference needed]. Figure 3 The detailed description of the embodiments disclosed herein will not be repeated here.

[0133] Step 604: Based on the network quality level and the user category, the trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0134] In some embodiments, the trained large model has the ability to generate content adapted to specific user categories. Before invocation, based on the user category determined in step 602, the corresponding trained large model is matched from the model library (e.g., if the user category is "technology enthusiast", then the trained large model adapted to the technology field is invoked). The network quality level obtained in step 601, the user category obtained in step 602, and the preprocessed summary obtained in step 603 are integrated into the model input according to a preset format (e.g., "Network quality level: [level]; User category: [category]; Preprocessed summary: [content]"). After receiving the input, the trained large model adjusts the expression style of the preprocessed summary based on the user category preferences it has learned (e.g., "technology enthusiast" corresponds to a style rich in professional terminology), and adjusts the amount of content data in combination with the network quality level (e.g., low level corresponds to concise text without images, high level corresponds to detailed text with images), generating personalized recommendation content.

[0135] By using the above methods, the trained large model can make full use of its learned user category adaptation capabilities, so that the generated personalized recommendation content matches user needs in terms of style and focus, while taking into account network quality adaptation, thereby further improving the adaptability of personalized recommendation content to users.

[0136] Figure 9 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further shown. Based on Figure 8 The illustrated embodiment further explains step 604. Figure 9 This may include the following steps: Step 701: Obtain the geographical location information of the terminal.

[0137] In some embodiments, the terminal's geographic location information refers to data reflecting the terminal's physical location, including but not limited to latitude and longitude coordinates (such as 30°N, 120°E) and administrative region identifiers (such as "Chaoyang District, Beijing" and "Shenzhen City, Guangdong Province"). Acquisition methods include the terminal actively providing the information and the system interface reading it: the terminal collects location information through its built-in positioning module (such as the Global Positioning System or the BeiDou Navigation Satellite System) and reports it to the system at a preset frequency (such as once every 30 seconds); or the system calls the terminal's positioning interface (such as the LocationManager interface in Android or the CoreLocation framework in iOS) to read the terminal's geographic location information. The acquired geographic location information is standardized in format and uniformly converted into administrative region identifiers or latitude and longitude coordinates for easy subsequent region matching.

[0138] Step 702: Call the large model of the corresponding regional edge node based on the geographical location information.

[0139] In some embodiments, edge nodes are divided by geographical region, with the division rules corresponding to administrative levels (e.g., provincial, municipal, district). Each edge node covers a specific region and deploys at least one large model (including the trained large model), and stores a "region-edge node" mapping table. When a large model is invoked based on geographic location information, the geographic location information is parsed to determine the region, and the mapping table is queried to find the corresponding edge node. A call request is sent to the edge node via a low-latency communication protocol (e.g., 5G edge computing protocol, near-field wireless communication protocol), and the request contains data such as preprocessed summary, network quality level, and user category. After receiving the call request, the edge node's large model processes the data according to preset logic, generates personalized recommendation content, and sends it to the terminal.

[0140] By using the above method, the nearest edge node large model is invoked based on the terminal's geographical location information, which reduces the transmission distance and time of data between the remote server and the terminal, reduces the generation delay of personalized recommendation content, and improves the real-time access of personalized recommendation content to end users.

[0141] Figure 10 A flowchart illustrating a content recommendation method provided in an embodiment of this disclosure is further shown. Based on Figure 3 The illustrated embodiment further explains step 102. Figure 10 This may include the following steps: Step 801: Obtain the user's feature data.

[0142] In some embodiments, user characteristic data is a multi-dimensional data set reflecting user attributes, behaviors, and preferences, including but not limited to basic attribute data (such as age, gender, and region), behavioral data (such as content browsing history, click timestamps, and dwell time), and preference data (such as content tags that users follow regularly and the frequency of search keywords). Acquisition methods include, but are not limited to: automatically collecting behavioral data generated by users during use through system logs; collecting basic attribute data actively provided by users through user registration or settings pages; and extracting preference-related data through content interaction records.

[0143] Step 802: Clean the feature data to remove outliers and duplicate data, and obtain cleaned feature data.

[0144] In some embodiments, outliers refer to feature data that deviates from the normal data distribution range. Identification methods include, but are not limited to, the standard deviation method (e.g., values ​​exceeding the mean ± 3 standard deviations) and the box plot method (e.g., values ​​exceeding 1.5 times the interquartile range). Duplicate data refers to records with completely identical content. Duplicate data is identified by comparing unique identifiers in the feature data. The cleaned feature data must maintain logical consistency; for example, ensuring that the "age" field is within a reasonable numerical range (e.g., 0-120 years old) and avoiding contradictory data.

[0145] Step 803: Process the cleaned feature data to obtain standardized feature data.

[0146] In some embodiments, the purpose of the processing is to convert feature data with different dimensions and value ranges into a unified scale to facilitate subsequent analysis. Processing methods include, but are not limited to, min-max standardization and Z-score standardization.

[0147] Step 804: Extract key features from the standardized feature data based on preset feature filtering rules.

[0148] In some embodiments, preset feature selection rules are used to retain features that significantly affect user category classification and exclude redundant or low-association features. Feature selection rules include, but are not limited to: variance thresholding (retaining features with variance higher than a preset threshold, such as variance > 0.1), excluding features with small value variations (such as "region" being entirely the same city); and feature importance ranking (calculating the contribution of features to the classification results using machine learning models (such as random forest, XGBoost) and selecting features based on their contribution).

[0149] Step 805: Analyze the key features to obtain at least one clustering result.

[0150] In some embodiments, the analysis method is a clustering algorithm, including but not limited to K-means clustering (presets the number of clusters K, such as K=5, grouping users with high similarity in key features into one category) and DBSCAN clustering (automatically divides the number of clusters based on density, grouping users who are close together into one category). During the analysis, by calculating the similarity between users' key feature vectors (such as cosine similarity, Euclidean distance), users with similar feature vectors are grouped into the same cluster, and each cluster represents a group of users with similar characteristics.

[0151] Step 806: Determine the user category corresponding to the user based on the at least one clustering result.

[0152] In some embodiments, clustering results are mapped to preset user categories. Mapping methods include, but are not limited to: assigning category labels to each cluster (e.g., "High-Frequency Beauty Browsing Group" corresponds to "Users with Deep Interest in Beauty"); establishing a correspondence between cluster centers and category labels by labeling typical user samples. The user category is determined based on the label of the cluster to which the user belongs. For example, if a user belongs to the "High-Frequency Beauty Browsing Group," then their user category is determined to be "Users with Deep Interest in Beauty."

[0153] By using the methods described above, the identified user categories can better reflect the actual characteristics and preferences of users, providing a precise user profile basis for subsequent personalized content recommendations.

[0154] Corresponding to the content recommendation method described above, this invention also proposes a content recommendation device. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0155] Figure 11 This is a schematic diagram of the structure of a content recommendation device provided in an embodiment of this disclosure, as shown below. Figure 11 As shown, the device is configured in a content recommendation system and includes: The first determining unit 91 is used to determine the network quality level of the terminal based on the acquired network quality information of the terminal; The second determining unit 92 is used to acquire the user's feature data, analyze the feature data, and determine the user category corresponding to the user. The first generation unit 93 is used to generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended. The second generation unit 94 is used to process the preprocessed summary based on the network quality level and the user category to generate personalized recommendation content.

[0156] The content recommendation device described in this embodiment can determine the network quality level based on terminal network quality information, determine the user category based on user feature data, generate a preprocessed summary according to the type of content to be recommended, and then combine the network quality level and user category to process the preprocessed summary to generate personalized recommendation content. This makes the generated personalized recommendation content both adaptable to terminal network conditions and meets user needs, thereby improving the adaptability of personalized recommendation content.

[0157] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the first determining unit 91 includes: The first acquisition module 911 is used to acquire the network transmission parameters and wireless signal parameters of the terminal. The first determining module 912 is used to process the network transmission parameters and the wireless signal parameters to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters. The second determining module 913 is used to calculate the standardized score based on a preset weight allocation rule to obtain a comprehensive score; The third determining module 914 is used to determine the network quality level of the terminal based on the comprehensive score.

[0158] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the first generation unit 93 includes: The fourth determining module 931 is used to extract text and split sentences from the content to be recommended to obtain a list of sentences; The fifth determining module 932 is used to determine the text importance score of each sentence in the sentence list; The judgment module 933 is used to determine whether the content to be recommended contains an image; The sixth determining module 934 is used to determine the relevance parameters between each sentence in the sentence list and the image when the content to be recommended contains the image, select the target sentence based on the relevance parameters and the text importance score, and generate the preprocessed summary; The seventh determining module 935 is used to select a target sentence based on the text importance score and generate the preprocessed summary when the content to be recommended does not contain the image.

[0159] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the second generation unit 94 includes: The generation module 941 is used to process the preprocessed summary by calling a large model based on the network quality level and the user category, and generate the personalized recommendation content.

[0160] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the device further includes: The eighth determining module 942 is used to construct a training dataset and train the large model after the generation module 941 processes the preprocessed summary based on the network quality level and the user category, thereby obtaining the trained large model.

[0161] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the generation module 941 is further configured to: Based on the network quality level and the user category, the trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

[0162] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the generation module 941 includes: The acquisition submodule 9411 is used to acquire the geographical location information of the terminal; Submodule 9412 is invoked to invoke the large model of the corresponding regional edge node based on the geographic location information.

[0163] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 12 As shown, the second determining unit 92 is further configured to: Obtain the user's feature data; The feature data is cleaned to remove outliers and duplicate data, resulting in cleaned feature data. The cleaned feature data is processed to obtain standardized feature data; Based on preset feature selection rules, key features are extracted from the standardized feature data; Analyze the key features to obtain at least one clustering result; Based on the at least one clustering result, determine the user category corresponding to the user.

[0164] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0165] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0166] Figure 13 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0167] like Figure 13As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in ROM (Read-Only Memory) 1002 or loaded from storage unit 1008 into RAM (Random Access Memory) 1003. The RAM 1003 may also store various programs and data required for the operation of the electronic device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. An I / O (Input / Output) interface 1005 is also connected to bus 1004.

[0168] Multiple components in electronic device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of displays, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows electronic device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0169] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as content recommendation methods. For example, in some embodiments, the content recommendation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the aforementioned content recommendation method by any other suitable means (e.g., by means of firmware).

[0170] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0171] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0172] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0173] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0174] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0175] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0176] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0177] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0178] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A content recommendation system, characterized in that, The system includes: a network quality assessment module, a user segmentation module, a content preprocessing module, and a summary generation module; wherein... The network quality assessment module is used to determine the network quality level of the terminal based on the obtained network quality information of the terminal, and send the network quality level to the summary generation module; The user segmentation module is used to analyze the acquired user feature data to determine the user category, and send the user category to the summary generation module. The content preprocessing module is used to perform differentiated processing based on the type of content to be recommended to generate a preprocessed summary, and send the preprocessed summary to the summary generation module; The summary generation module is used to receive the network quality level, the user category, and the preprocessed summary, and to call a large model to process the preprocessed summary according to the network quality level and the user category to generate personalized recommendation content; wherein, after receiving the input data, the large model parses the network quality level to determine the degree of conciseness of the generated personalized recommendation content; and parses the user category to adjust the content style, adopting a corresponding language style for different user categories; The content preprocessing module includes: a text processing unit, a matching unit, a scoring unit, and a summary generation unit; wherein, The text processing unit is used to extract text and split sentences into the content to be recommended, obtain a sentence list, and send the sentence list to the matching unit and / or the scoring unit. The matching unit is configured to receive the sentence list when the content to be recommended contains an image, determine the relevance parameters between each sentence in the sentence list and the image, and send the relevance parameters to the summary generation unit. The scoring unit is used to receive the sentence list, determine the text importance score of each sentence in the sentence list, and send the text importance score to the summary generation unit; The summary generation unit is used to select target sentences and generate the preprocessed summary based on the relevance parameter and the text importance score when the content to be recommended contains the image. The summary generation unit is further configured to generate the preprocessed summary by selecting target sentences based on the text importance score when the content to be recommended does not contain the image.

2. The system according to claim 1, characterized in that, The network quality assessment module includes: an acquisition unit, a processing unit, and a determination unit; wherein, The acquisition unit is used to acquire the network transmission parameters and wireless signal parameters of the terminal; and send the network transmission parameters and wireless signal parameters to the processing unit. The processing unit is configured to process the network transmission parameters and the wireless signal parameters to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters; and send the standardized scores to the determining unit. The determining unit is used to calculate the standardized scores corresponding to the network transmission parameters and the wireless signal parameters based on a preset weight allocation rule, obtain a comprehensive score, and determine the network quality level of the terminal based on the comprehensive score.

3. The system according to claim 1, characterized in that, The system also includes: an adjustment module; The adjustment module is used to construct a training dataset, train the large model called by the summary generation module to obtain the trained large model, and send the trained large model to the summary generation module.

4. The system according to claim 3, characterized in that, The summary generation module is also used for: The trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

5. The system according to claim 4, characterized in that, The system also includes: an edge deployment module; The edge deployment module is used to call the large model of the edge nodes in the corresponding area based on the geographical location information of the terminal.

6. A content recommendation method, characterized in that, The content recommendation method is applied to the content recommendation system according to any one of claims 1-5, comprising: The network quality level of the terminal is determined based on the obtained network quality information of the terminal; Obtain user feature data, analyze the feature data, and determine the user category corresponding to the user; Based on the type of content to be recommended, a preprocessed summary is generated through differentiated processing. Based on the network quality level and the user category, a large model is invoked to process the preprocessed summary and generate personalized recommendation content. The large model receives input data, parses the network quality level to determine the conciseness of the generated personalized recommendation content, and parses the user category to adjust the content style, adopting corresponding language styles for different user categories. The step of generating a preprocessed summary by performing differentiated processing based on the type of content to be recommended includes: The text to be recommended is extracted and the sentences are split to obtain a list of sentences; Determine the text importance score for each sentence in the sentence list; Determine whether the content to be recommended contains images; If the content to be recommended contains the image, then the relevance parameters between each sentence in the sentence list and the image are determined, and the target sentence is selected based on the relevance parameters and the text importance score to generate the preprocessed summary; If the content to be recommended does not contain the image, then the target sentence is selected based on the text importance score to generate the preprocessed summary.

7. The method according to claim 6, characterized in that, Determining the network quality level of the terminal based on the acquired network quality information of the terminal includes: Obtain the network transmission parameters and wireless signal parameters of the terminal; The network transmission parameters and the wireless signal parameters are processed to obtain standardized scores corresponding to the network transmission parameters and the wireless signal parameters; The standardized scores are calculated based on preset weighting rules to obtain a comprehensive score; The network quality level of the terminal is determined based on the comprehensive score.

8. The method according to claim 6, characterized in that, After processing the preprocessed summary using a large model based on the network quality level and the user category to generate the personalized recommendation content, the method further includes: A training dataset is constructed, and the large model is trained to obtain the trained large model.

9. The method according to claim 8, characterized in that, The process of processing the preprocessed summary based on the network quality level and the user category, and generating the personalized recommendation content by calling a large model, includes: Based on the network quality level and the user category, the trained large model is invoked to process the preprocessed summary and generate the personalized recommendation content.

10. The method according to claim 6, characterized in that, The process of processing the preprocessed summary based on the network quality level and the user category, and generating the personalized recommendation content by calling a large model, includes: Obtain the geographical location information of the terminal; The large model of the corresponding regional edge node is invoked based on the geographical location information.

11. The method according to claim 6, characterized in that, The process of acquiring user feature data, analyzing the feature data, and determining the user category corresponding to the user includes: Obtain the user's feature data; The feature data is cleaned to remove outliers and duplicate data, resulting in cleaned feature data. The cleaned feature data is processed to obtain standardized feature data; Based on preset feature selection rules, key features are extracted from the standardized feature data; Analyze the key features to obtain at least one clustering result; Based on the at least one clustering result, determine the user category corresponding to the user.

12. A content recommendation device, characterized in that, The content recommendation device is configured in the content recommendation system according to any one of claims 1-5, comprising: The first determining unit is used to determine the network quality level of the terminal based on the network quality information of the acquired terminal. The second determining unit is used to acquire the user's feature data, analyze the feature data, and determine the user category corresponding to the user. The first generation unit is used to generate a preprocessed summary by performing differentiated processing based on the type of content to be recommended. The second generation unit is used to process the preprocessed summary by calling a large model based on the network quality level and the user category to generate personalized recommendation content; wherein, after receiving the input data, the large model parses the network quality level to determine the conciseness of the generated personalized recommendation content; and parses the user category to adjust the content style, adopting a corresponding language style for different user categories; The first generation unit includes: The fourth determining module is used to extract text and split sentences from the content to be recommended to obtain a list of sentences; The fifth determining module is used to determine the text importance score of each sentence in the sentence list; The judgment module is used to determine whether the content to be recommended contains images; The sixth determining module is used to determine the relevance parameters between each sentence in the sentence list and the image when the content to be recommended contains the image, select the target sentence based on the relevance parameters and the text importance score, and generate the preprocessed summary; The seventh determining module is used to select a target sentence based on the text importance score and generate the preprocessed summary when the content to be recommended does not contain the image.

13. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 6-11.

14. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 6-11.

15. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 6-11.

Images