Content recommendation method and device
By constructing a hot word association network and analyzing user profiles, a personalized content recommendation list is generated, which solves the problems of homogeneity and poor timeliness in existing technologies, and achieves more accurate and diversified recommendation results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN ZHONGJIERUN TECH CO LTD
- Filing Date
- 2026-01-07
- Publication Date
- 2026-04-17
AI Technical Summary
Existing content recommendation methods cannot delve into the deep semantic relationships and dynamic changes of content, resulting in highly homogenized recommendation results and poor timeliness, making it difficult to meet users' personalized and diverse content needs.
By constructing a hot word association network, based on real-time hot word data and user profile analysis, hot word association mapping is performed to generate a personalized content recommendation list. The matching degree value is calculated by combining the hierarchical information and frequency of user interest hot words in the hot word association network.
It achieves precise correlation between content features and user interest features, improving the accuracy and diversity of recommendations, dynamically capturing content changes and user interests, and avoiding problems of homogenization and poor timeliness.
Smart Images
Figure CN121880653A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer technology, and in particular to a content recommendation method and apparatus. Background Technology
[0002] With the booming development of the internet industry, the amount of content has exploded, covering multiple fields such as games, entertainment, education, and e-commerce. Users struggle to quickly find content that matches their interests and needs amidst this massive amount of information, leading to a decline in user experience and impacting the dissemination efficiency of high-quality content. To address this issue, various content recommendation methods have emerged. Their core purpose is to analyze user behavior data and content characteristics to accurately push potentially interesting content to users, increasing user viewing time and satisfaction, and promoting the healthy development of the platform.
[0003] However, the feature mining of content during the recommendation process is not deep enough, often relying solely on single-dimensional keywords or simple content classifications for matching, failing to effectively capture the deep semantic relationships and dynamic trends of the content. Furthermore, when combining user information, it fails to accurately and dynamically associate user interests with the complex features of the content, resulting in highly homogenized and untimely recommendation results that are difficult to meet users' personalized and diverse content needs. Summary of the Invention
[0004] This invention provides a content recommendation method and apparatus to improve the accuracy and diversity of content recommendations and avoid homogenization and poor timeliness.
[0005] In a first aspect, the present invention provides a content recommendation method, including: Collect real-time hot word data for each piece of content currently being published on the platform; the real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; A hot word association network is constructed based on real-time hot word data for each content. Each node in the hot word association network represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. User profile analysis is performed based on user profile data to obtain user interest hot words; the user profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records. Based on the hot word association network of each content and the hot words of user interest, an association mapping is performed to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; Based on the hot word association mapping results for each content, the matching degree value between each content and the user's interest hot words is determined, and a personalized content recommendation list is generated based on the matching degree value of each content.
[0006] In a second aspect, the present invention also provides a content recommendation apparatus, applied to the content recommendation method as described in the first aspect; the content recommendation apparatus includes: The data collection module is used to collect real-time hot word data for each piece of content currently being published on the platform; the real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; The hot word relationship construction module is used to build a hot word association network based on real-time hot word data for each content. Each node in the hot word association network represents the hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. The user profile analysis module is used to perform user profile analysis based on user profile data to obtain user interest hot words; the user profile data includes the user's historical viewing records, historical interaction records, and interest tags in the registration records. The hot word association mapping module is used to perform association mapping based on the hot word association network of each content and the hot words of user interest to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; The content recommendation module is used to determine the matching degree between each content and the user's interest hot words based on the hot word association mapping results of each content, and to generate a personalized content recommendation list based on the matching degree value of each content.
[0007] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby implementing the method recommended above.
[0008] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the method recommended as described above.
[0009] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method recommended as described above.
[0010] The content recommendation method provided in this invention constructs a hot word association network based on real-time hot word data for each content. By dividing hot words into levels and frequencies of occurrence, it deeply mines the deep semantic relationships and features of the content. Through user profile analysis of user profile data, it accurately captures user interest hot words. Then, it uses the constructed hot word association network and user interest hot words for association mapping. Combining the number of nodes appearing in the hot word association network for user interest hot words, as well as the level information and frequency of each node in the hot word association network, it achieves accurate association between content features and user interest features. Finally, based on the matching degree value between the content and user interest hot words determined by the hot word association mapping results, it generates a personalized content recommendation list. This ensures that the recommendation results meet the personalized needs of users and can dynamically capture changes in content and user interests, improving the accuracy and diversity of recommendations and avoiding homogenization and poor timeliness. Attached Figure Description
[0011] Figure 1 This is a flowchart illustrating the content recommendation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the content recommendation device provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation
[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0014] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the present invention. In the following description, details are set forth for purposes of explanation. It should be understood that the present invention may be implemented without these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.
[0015] Referring to Figure 1 , Figure 1 is a schematic flowchart of the content recommendation method provided by the present invention. In the embodiments of the present invention, the execution subject of the content recommendation method is a recommendation device. Therefore, the content recommendation method includes: Step 10, collecting the real-time hot word data of each content being carried out on the current platform. The real-time hot word data includes hot word texts and the occurrence frequencies of each hot word text.
[0016] Optionally, the recommendation device obtains in real time the text information in the audio stream and video stream of each content being carried out on the current platform. Among them, the audio stream needs to be converted into text through speech-to-text technology, and the video stream needs to extract text contents such as bullet screens and subtitles. <8000058>Further, the recommendation device preprocesses the collected text, including removing punctuation marks and stop words (such as meaningless words like "de" and "le"), and performing word segmentation. Further, the recommendation device counts the occurrence frequencies of each word after word segmentation, screens out the words with higher frequencies as the real-time hot words of the content, and obtains the real-time hot word data including hot word texts and corresponding occurrence frequencies.
[0018] In one embodiment, a "Makeup Tutorial for XX Brand" is underway on the platform. The audio stream is acquired in real-time, and text such as "Today I'm recommending this foundation; the texture is very light and the coverage is good..." is obtained through speech-to-text technology. Simultaneously, the video's comment text, such as "I've used this foundation, it's super good!" and "Please provide a link to the foundation," is extracted. These texts are preprocessed, removing punctuation and stop words like "today" and "to everyone," and then segmented into words such as "foundation," "texture," "lightweight," "coverage effect," "good," and "link." Statistics show that "foundation" appeared 8 times, "good" appeared 5 times, and "coverage effect" appeared 3 times. Therefore, the real-time hot word data is as follows: hot word text "foundation," frequency 8 times; hot word text "good," frequency 5 times; hot word text "coverage effect," frequency 3 times.
[0019] Step 20: Construct a hot word association network based on the real-time hot word data for each content. In the hot word association network, each node represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence.
[0020] Furthermore, the recommendation device performs semantic analysis on the hot words to clarify the hierarchical relationships between them. Generally, broader hot words are designated as upper-level nodes, while more specific hot words are designated as lower-level nodes. The recommendation device then presents each hot word text as a node, with hierarchical relationships between nodes expressed through connections. Simultaneously, the frequency of hot word occurrences is used as an attribute of each node, thus completing the construction of the hot word association network, as detailed in steps 201 to 204.
[0021] Step 30: Perform user profile analysis based on user profile data to obtain user interest hot keywords. User profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records.
[0022] Furthermore, the recommendation system collects user profile data, which includes interest tags from users' historical viewing records, historical interaction records, and registration records. Historical viewing records are analyzed to statistically analyze the types of content viewed by users and related trending words; historical interaction records, such as likes, comments, shares, and purchases, are summarized, and combined with interest tags from registration records, this information is analyzed to extract words that users frequently use and that have high user interest, identifying these as user interest trending words.
[0023] In one embodiment, user A's user profile data shows that their viewing history indicates they have repeatedly watched beauty-related content, especially tutorials on base makeup; their interaction history shows they have repeatedly liked and commented on content related to "foundation" and "concealing effect" in the beauty category, and they have previously purchased a certain brand of foundation; their registration records show their interest tags as "beauty" and "base makeup." Analyzing this data, user A frequently uses terms such as "beauty," "base makeup," "foundation," and "concealing effect," and shows a high level of interest in these terms. Therefore, user A's top interest keywords are determined to be "beauty," "base makeup," "foundation," and "concealing effect."
[0024] Step 40: Based on the hot word association network and user interest hot words for each content, perform association mapping to obtain the hot word association mapping results. The hot word association mapping results include the number of nodes in the hot word association network where user interest hot words appear, as well as the hierarchical information and frequency of each node in the hot word association network.
[0025] Furthermore, the recommendation device maps the hot word association network of each content to user interest hot words, checking one by one whether a corresponding node exists in the hot word association network for each user interest hot word. If it does, the number of such nodes is recorded. For each existing node, its hierarchical information (i.e., the hierarchical position of the node) and the frequency of its occurrence in the hot word association network are also recorded. This ultimately forms a hot word association mapping result that includes the number of nodes in the hot word association network for user interest hot words, as well as the hierarchical information and frequency of each occurrence node in the hot word association network.
[0026] Continuing with the above example, using the hot word association network example of user A's interest keywords "makeup," "base makeup," "foundation," "concealing effect," and "XX brand makeup tutorial," the recommendation device performs association mapping and finds that "makeup" corresponds to the root node (level 1) in the tree structure, with a frequency of 10; "base makeup" corresponds to the "base makeup product" node (level 2), with a frequency of 7; "foundation" corresponds to the "foundation" node (level 3), with a frequency of 8; and "concealing effect" corresponds to the "concealing effect" node (level 4), with a frequency of 3. The number of nodes for this user's interest keywords in this hot word association network is 4. Therefore, the hot word association mapping result is: 4 nodes; "makeup" node level 1, frequency 10; "base makeup product" node level 2, frequency 7; "foundation" node level 3, frequency 8; and "concealing effect" node level 4, frequency 3.
[0027] Step 50: Determine the matching degree value between each content and the user's interest hot words based on the hot word association mapping results of each content, and generate a personalized content recommendation list based on the matching degree value of each content.
[0028] Furthermore, the recommendation device calculates a matching degree value based on the hot word association mapping results for each content. Different weights can be set; generally, the shallower the node level (closer to the root node), the lower the weight, and the deeper the level, the higher the weight. Higher frequency of occurrence also results in higher weight. Optionally, in this embodiment of the invention, the matching degree value between the content and the user's interest hot words is obtained by multiplying the level weight and frequency weight of each matching node and then summing the results. Further, the recommendation device sorts all the matching degree values and generates a personalized content recommendation list in descending order.
[0029] In one embodiment, the weight of level 1 is set to 0.2, level 2 to 0.3, level 3 to 0.4, and level 4 to 0.5. The frequency weight is the frequency of occurrence divided by the highest frequency in the hot word association network (in this example, the highest frequency is 10). For "XX brand beauty tutorial", the "beauty" node has a weight of 0.2*(10 / 10)=0.2; the "base makeup product" node has a weight of 0.3*(7 / 10)=0.21; the "foundation" node has a weight of 0.4*(8 / 10)=0.32; and the "concealing effect" node has a weight of 0.5*(3 / 10)=0.15. Therefore, the matching degree is 0.2+0.21+0.32+0.15=0.88.
[0030] Suppose there's another event on the platform called "YY Brand Skincare," which has a match score of 0.65 with user A's trending topics. Then, in the personalized content recommendation list generated for user A, sorted from highest to lowest match score, "XX Brand Makeup Tutorial" will appear before "YY Brand Skincare."
[0031] This invention constructs a hot word association network based on real-time hot word data for each content. By dividing hot words into levels and frequencies, it delves into the deep semantic relationships and features of the content. Through user profile analysis of user profile data, it accurately captures user interest hot words. Then, it uses the constructed hot word association network and user interest hot words for association mapping. Combining the number of nodes in the hot word association network for user interest hot words, as well as the level information and frequency of each node in the hot word association network, it achieves a precise association between content features and user interest features. Finally, based on the matching degree value between the content and user interest hot words determined by the hot word association mapping results, it generates a personalized content recommendation list. This ensures that the recommendation results meet the personalized needs of users and can dynamically capture content changes and user interests, improving the accuracy and diversity of recommendations and avoiding homogenization and poor timeliness.
[0032] In one embodiment, steps 201 to 204 include: Step 201: For any two first hot word texts and second hot word texts in the same content, if the text similarity between the first hot word text and the second hot word text is greater than or equal to the similarity threshold, then the text inclusion relationship between the first hot word text and the second hot word text is determined.
[0033] Optionally, the recommendation device selects any two hot word texts (i.e., the first hot word text and the second hot word text) from the real-time hot word data within the same content. A text similarity calculation algorithm is used to calculate the similarity between these two hot word texts. A similarity threshold is set; if the calculated similarity is greater than or equal to this threshold, it is determined that there is a textual inclusion relationship between the first and second hot word texts. The text similarity calculation uses a cosine similarity algorithm.
[0034] In one embodiment, among the real-time trending words for "XX brand beauty tutorials," the first trending word is selected as "base makeup products," and the second trending word is selected as "foundation." The recommendation device converts these two trending word texts into word vectors and calculates a similarity of 0.7 using the aforementioned cosine similarity formula. A similarity threshold of 0.6 is set. Since 0.7 > 0.6, it is determined that there is a textual inclusion relationship between "base makeup products" and "foundation."
[0035] Step 202: Determine the set of node relationships based on text inclusion relationships, and construct the same branch of the hot word association network for the same content based on hierarchical clustering combined with the set of node relationships.
[0036] Furthermore, the recommendation device determines the node relationship set based on the text inclusion relationship, as specifically in steps 2021 to 2024. The node relationship set includes the parent-child node relationship between hot word texts and the parent node compatible sibling node relationship.
[0037] Furthermore, the recommendation device employs a hierarchical clustering method, combining a set of node relationships to construct a hot word association network with the same content, forming branches. Within each branch, the hot word text without a parent node is used as the root node. Based on the parent-child relationships of the root node in the node relationship set, all direct child nodes are connected to the root node. For each node designated as the first target child node, all direct child nodes are connected to it based on its parent-child relationships in the node relationship set. For any two nodes designated as the second target nodes, parent-child connections and sibling connections are established respectively based on their parent-sibling relationships in the node relationship set.
[0038] Optionally, nodes within the same branch of each hot word association network form a complete hierarchical chain according to their subordinate relationships, denoted as the same branch of the hot word association network. ,in, This refers to the number of branches within the same branch of the hot word association network. Indicates the first Each branch of the hot word association network contains a hierarchical sequence of hot word texts.
[0039] Continuing with the above embodiments, in the "XX Brand Makeup Tutorial" node relationship set, "Makeup Product" is the parent node of "Base Makeup Product," "Base Makeup Product" is the parent node of "Foundation," "Foundation" is the parent node of "XX Brand Foundation," "Texture," "Lightweight," "Coverage Effect," and "Easy to Use," and "Texture" and "Lightweight" are sibling nodes of the parent node. The recommended device uses "Makeup Product," which has no parent node, as the root node, and connects its direct child node "Base Makeup Product" to the root node. "Base Makeup Product" is the first target child node, and connects its direct child node "Foundation" to itself. "Foundation" is the first target child node, and connects its direct child nodes "XX Brand Foundation," "Texture," "Lightweight," "Coverage Effect," and "Easy to Use" to itself. For the two second target nodes, "Texture" and "Lightweight," sibling node connections are established, and both are connected to "Foundation" as parent-child nodes, forming a hot word association network with the same branch.
[0040] Step 203: Construct cross-branch relationships of hot word association networks based on the same branch of hot word association networks with different content.
[0041] Furthermore, the recommendation device analyzes the same branches of the hot word association network for different content to find the association relationships between hot word texts in the same branches of different hot word association networks. These association relationships can be semantically similar, conceptually related, etc. Further, based on these association relationships, the recommendation device constructs cross-branch relationships in the hot word association network, clarifying the connection methods and degree of association between the same branches of the hot word association networks for different content, as detailed in steps 2301 to 2034.
[0042] Step 204: Construct a hot word association network based on the same branch and cross-branch relationships of the hot word association network.
[0043] Furthermore, the recommendation device integrates the same-branch and cross-branch relationships of the hot word association network for each content, arranges the same-branch hot word association network according to its internal hierarchical chain, and then connects the different same-branch hot word association networks through the cross-branch relationships of the hot word association network to obtain the hot word association network, as specifically in steps 2041 to 2044. The hot word association network can comprehensively reflect the hierarchical subordinate relationship and cross-content association relationship between the hot words of each content.
[0044] The hot word association network constructed in this embodiment of the invention can clearly present the hierarchical relationship between hot words of various content, including the detailed classification and subordinate connection of hot words within the same content. At the same time, it can also reflect the relationship between hot words of different content, providing accurate hot word relationships for subsequent content recommendation based on user interest hot words. Therefore, it can more accurately find content that matches user interests and improve the accuracy and rationality of personalized content recommendation.
[0045] In one embodiment, steps 2021 to 2024 include: Step 2021: If the semantic inclusion relationship is that the first semantic range of the first hot word text completely includes the second semantic range of the second hot word text, then determine that the first hot word text is the first parent-child node of the second hot word text.
[0046] Optionally, the recommendation device further analyzes the semantic scope of a first hot word text and a second hot word text that have a textual inclusion relationship within the same content. Through semantic analysis technology, the first semantic scope of the first hot word text and the second semantic scope of the second hot word text are clarified. If the analysis determines that the first semantic scope completely includes the second semantic scope, then the first hot word text is determined to be the parent node of the second hot word text, thus forming a first parent-child node relationship.
[0047] In one embodiment, in the "XX Brand Makeup Tutorial," the first hot keyword text is "base makeup products," and the second hot keyword text is "foundation." The recommendation device performs semantic analysis on both. The first semantic scope of "base makeup products" covers all makeup products used for facial base application, including foundation, cushion foundation, BB cream, etc.; the second semantic scope of "foundation" refers only to liquid base makeup products that have functions such as concealing blemishes and correcting skin tone. Obviously, the first semantic scope completely includes the second semantic scope, therefore, "base makeup products" is determined to be the first parent-child node of "foundation."
[0048] Step 2022: If the semantic inclusion relationship is that the second semantic scope completely includes the first semantic scope, then determine that the second hot word text is the second parent-child node relationship of the first hot word text.
[0049] Furthermore, the recommendation device also analyzes the semantic scope of the first and second hot word texts that have a text inclusion relationship. When the analysis results show that the second semantic scope of the second hot word text completely includes the first semantic scope of the first hot word text, the second hot word text is determined to be the parent node of the first hot word text, forming a second parent-child node relationship.
[0050] Continuing with this, the first hot keyword text is "XX brand foundation liquid," and the second hot keyword text is simply "foundation liquid." Recommendation analysis reveals that the first semantic scope of "XX brand foundation liquid" refers only to foundation liquid products from a specific brand; the second semantic scope of "foundation liquid" covers foundation liquid products from all brands. Since the second semantic scope completely encompasses the first semantic scope, "foundation liquid" is determined to be the second parent-child node of "XX brand foundation liquid."
[0051] Step 2023: If the semantic inclusion relationship is that the first semantic range does not completely include the second semantic range, then determine that the first hot word text and the second hot word text are parent node compatible sibling nodes.
[0052] Furthermore, for the first hot word text and the second hot word text that have a text inclusion relationship, after the recommendation device analyzes their semantic range, if it finds that the first semantic range of the first hot word text does not completely include the second semantic range of the second hot word text, and the second semantic range does not completely include the first semantic range, that is, the semantic ranges of the two are partially related but do not completely include each other, then the first hot word text and the second hot word text are determined to be parent node compatible sibling nodes.
[0053] Continuing with the hot words, the first hot word text is "texture," and the second hot word text is "light and thin." The recommendation device analyzes the semantic scope. The first semantic scope of "texture" refers to the properties and state of an object, including the fineness and consistency of a product; the second semantic scope of "light and thin" refers to the product's light and non-heavy texture. The first semantic scope does not completely encompass the second semantic scope; they are semantically related but not mutually exclusive. Therefore, "texture" and "light and thin" are determined to be parent nodes with compatible sibling relationships.
[0054] Step 2024: Determine the first parent-child node relationship, the second parent-child node relationship, and the parent node compatible sibling node relationship as a set of node relationships.
[0055] Furthermore, the recommendation device summarizes the first parent-child node relationship, the second parent-child node relationship, and the parent node compatible sibling node relationship, integrating different types of node relationships to obtain a node relationship set. The node relationship set covers various types of node relationships between hot word texts in the same content.
[0056] Continuing with the "XX Brand Makeup Tutorial," the previous steps established the first parent-child node relationship between "Base Makeup Product" and "Foundation," the second parent-child node relationship between "Foundation" and "XX Brand Foundation," the parent node compatibility sibling relationship between "Texture" and "Lightweight," and the first parent-child node relationships between "Foundation" and "Texture," "Foundation" and "Lightweight," etc. The recommendation device summarizes these relationships to form the node relationship set.
[0057] The embodiments of the present invention can accurately determine the node relationships between various hot words in the same content, forming a complete set of node relationships. This provides a clear basis for the subsequent construction of the hot word association network within the same branch, ensuring the accuracy and logic of node connections within the hot word association network within the same branch. This enables the hot word association network to more accurately reflect the hierarchical subordinate and association situation between hot words, laying a reliable relational foundation for subsequent stages of personalized content recommendation.
[0058] In one embodiment, steps 2031 to 2034 include: Step 2031: For any two branches of the first hot word association network of the first target content and the second hot word association network of the second target content, determine the semantic similarity between the first target hot word text and the second target hot word text based on the first target hot word text in the first hot word association network branch and the second target hot word text in the second hot word association network branch.
[0059] Optionally, the recommendation device selects any two different pieces of content, namely the first target content and the second target content, and extracts the same branch of their hot word association network, denoted as the first hot word association network branch and the second hot word association network branch. The first target hot word text is selected from the first hot word association network branch, and the second target hot word text is selected from the second hot word association network branch. These two target hot word texts are analyzed using a semantic similarity calculation method to determine their semantic similarity. Semantic similarity is used to measure the degree of semantic connection between the two hot word texts.
[0060] In one embodiment, the first target content is "XX brand beauty tutorial," and the first target hot word text in the same branch of its first hot word association network is "beauty products"; the second target content is "ZZ brand makeup recommendation," and the second target hot word text in the same branch of its second hot word association network is "makeup products." The recommendation device calculates semantic similarity and finds that the semantic similarity between "beauty products" and "makeup products" is 0.8.
[0061] Step 2032: Based on semantic similarity and the first occurrence frequency of the first target hot word text and the second occurrence frequency of the second target hot word text, determine the semantic association strength between the first target hot word text and the second target hot word text.
[0062] Furthermore, the recommendation device calculates the semantic association strength between the first target hot word text and the second target hot word text based on the semantic similarity between them, combined with the first occurrence frequency of the first target hot word text and the second occurrence frequency of the second target hot word text, according to a preset calculation formula.
[0063] The formula for calculating semantic association strength is: r(a,b)=s(a,b)*log[f(a)+f(b)+1].
[0064] Where r(a,b) represents the semantic association strength between the first target hot word text h(a) and the second target hot word text h(b), s(a,b) represents the semantic similarity between the first target hot word text h(a) and the second target hot word text h(b), f(a) represents the first occurrence frequency of the first target hot word text h(a), f(b) represents the second occurrence frequency of the second target hot word text h(b), and log() represents the logarithmic function.
[0065] Continuing with the above example, the first occurrence frequency of "beauty product" f(a) = 10, the second occurrence frequency of "makeup product" f(b) = 8, and the semantic similarity s(a,b) = 0.8. The semantic association strength is calculated according to the formula: r(a,b) = 0.8 * log(10 + 8 + 1) = 0.8 * log(19) ≈ 0.8 * 1.2788 = 1.023.
[0066] Step 2033: The maximum semantic association strength between hot word texts is determined as the branch association strength between the same branch of the first hot word association network and the same branch of the second hot word association network.
[0067] Furthermore, the recommendation device pairs all hot word texts in the same branch of the first hot word association network with all hot word texts in the same branch of the second hot word association network, and calculates the semantic association strength between each pair of hot word texts according to the method in step 2032. Further, among all the calculated semantic association strengths, the recommendation device selects the largest value and determines it as the branch association strength between the same branch of the first hot word association network and the same branch of the second hot word association network.
[0068] In one embodiment, within the same branch of the first hot word association network for "XX Brand Makeup Tutorial" and the same branch of the second hot word association network for "ZZ Brand Makeup Recommendation," besides "makeup products" and "makeup products," there are also multiple pairs of hot word texts such as "foundation products" and "eyeshadow," and "foundation" and "makeup tools." The semantic association strength between "makeup products" and "makeup products" is calculated to be 1.023, while the semantic association strengths of the other pairs of hot word texts are all less than this value. Therefore, the branch association strength between the two branches is determined to be 1.023.
[0069] Step 2034: If the branch association strength is greater than or equal to the preset association strength threshold, and the root nodes of the same branch of the first hot word association network and the same branch of the second hot word association network have a semantic inclusion relationship, then it is determined that there is a cross-branch subordinate relationship between the same branch of the first hot word association network and the same branch of the second hot word association network. Based on the cross-branch subordinate relationship between the same branches of all hot word association networks, the cross-branch relationship of the hot word association network is determined.
[0070] Furthermore, a preset association strength threshold is set. If the branch association strength between the same branch of the first hot word association network and the same branch of the second hot word association network is greater than or equal to this threshold, and further analysis reveals that the root nodes of these two branches have a semantic inclusion relationship, the recommendation device determines that there is a cross-branch subordinate relationship between them. Further, the recommendation device summarizes the cross-branch subordinate relationships between the same branches of all hot word association networks to determine the cross-branch relationships of the hot word association networks.
[0071] In one embodiment, a preset association strength threshold of 0.8 is set. The root node of the first hot word association network for "XX Brand Makeup Tutorial" is "makeup products," and the root node of the second hot word association network for "ZZ Brand Makeup Recommendation" is "makeup products." It is known that the branch association strength between the two is 1.023 > 0.8, and the semantic scope of "makeup products" completely encompasses the semantic scope of "makeup products," meaning there is a semantic inclusion relationship between the root nodes. Therefore, a cross-branch subordinate relationship is determined between these two branches, and this relationship is included in the cross-branch relationship of the hot word association network.
[0072] The embodiments of this invention can accurately construct cross-branch relationships between hot word association networks of different content, clearly reflecting the degree of association and subordinate relationship between hot word systems of different content. This makes the constructed hot word association network not only include the hierarchical relationship of hot words within a single content, but also reflect the association between hot words of different content. This provides a comprehensive basis for subsequent accurate cross-content matching and personalized recommendations based on user interest hot words, improving the richness and accuracy of recommendations.
[0073] In one embodiment, steps 2041 to 2044 include: Step 2041: Based on the cross-branch subordination relationship in the hot word association network, construct a cross-content hierarchical structure. The hot word association network within the same content maintains the original hierarchical subordination relationship within the same branch. Add the corresponding occurrence frequency as attribute information to each node to obtain the initial structure tree.
[0074] Optionally, the recommendation device connects the same branches of hot word association networks for different content based on the cross-branch subordination relationships in the hot word association network, forming a cross-content hierarchical structure. For the same branch of the hot word association network within the same content, its original hierarchical subordination relationship remains unchanged. At the same time, the corresponding occurrence frequency is added as attribute information to each node in the initial structure tree, resulting in the initial structure tree. The connection relationships between nodes in the tree fully represent the subordination relationships between all hot word levels.
[0075] In one embodiment, the keyword association network of "XX Brand Makeup Tutorial" has a hierarchical relationship of "makeup products - base makeup products - foundation - each sub-node" within the same branch, and the keyword association network of "ZZ Brand Makeup Recommendation" has a hierarchical relationship of "makeup products - eyeshadow - eyeshadow texture" within the same branch, and the two have a cross-branch subordinate relationship. The recommendation device constructs a cross-content hierarchical structure based on this cross-branch subordinate relationship, placing the "makeup products" branch under the "makeup products" branch. Branches within the same content maintain their original hierarchy; for example, "base makeup products" still belongs to "makeup products," and "foundation" still belongs to "base makeup products," etc. A frequency attribute is added to each node, such as an attribute of 10 for the "makeup products" node and an attribute of 8 for the "makeup products" node, resulting in an initial structure tree.
[0076] Step 2042: Based on the sum of the frequencies of all nodes in each branch of the initial tree structure, and combined with the level depth of each branch in the initial tree structure, determine the importance index of each branch. The level depth of the root node is 1, and the depth increases by 1 for each level downward.
[0077] Furthermore, the recommendation device first calculates the sum of the occurrence frequencies of all nodes within each branch in the initial tree structure, and then determines the level depth of each branch in the initial tree structure, with the level depth of the root node being 1, and the depth increasing by 1 for each subsequent level. Further, the recommendation device combines these two data points to determine the importance index of each branch. Optionally, in this embodiment of the invention, the importance index is calculated using the formula: I = S / D, where I represents the branch importance index, S represents the sum of the occurrence frequencies of all nodes within the branch, and D represents the level depth of the branch.
[0078] In one embodiment, in the initial tree structure, the "Beauty Products" branch includes nodes "Beauty Products" (frequency 10), "Base Makeup Products" (frequency 7), "Foundation" (frequency 8), etc., with the sum of the frequencies of all nodes S = 10 + 7 + 8 + 6 + 2 + 4 + 3 + 5 = 45. The branch's level depth D = 1, and its importance index I = 45 / 1 = 45. The "Makeup Products" branch includes nodes "Makeup Products" (frequency 8), "Eyeshadow" (frequency 6), and "Eyeshadow Texture" (frequency 4), with the sum of the frequencies S = 8 + 6 + 4 = 18, level depth D = 2, and importance index I = 18 / 2 = 9.
[0079] Step 2043: The branches with importance indices greater than or equal to the preset index threshold are identified as core branches, and dynamic association paths are constructed based on the semantic similarity and semantic association strength of nodes in any two core branches.
[0080] Furthermore, a preset index threshold is set, and the recommendation device identifies branches with an importance index greater than or equal to this threshold as core branches. Further, the recommendation device performs pairwise analysis on nodes in any two core branches, calculating the semantic similarity and semantic association strength between the nodes (the semantic association strength calculation is the same as in step 2032), and connects nodes with semantic similarity greater than or equal to the preset similarity threshold and semantic association strength greater than or equal to the preset association strength threshold to construct dynamic association paths. These dynamic association paths are used to connect closely related nodes in different core branches.
[0081] In one embodiment, a preset index threshold of 10, a preset association strength threshold of 0.25, and a preset association strength threshold of 0.35 are set. The "beauty products" branch has an importance index of 45 > 10, and is therefore identified as a core branch. It is also assumed that the "skincare products" branch has an importance index of 15 > 10, and is also a core branch. The semantic similarity between the node "foundation" in the "beauty products" branch and the node "moisturizing essence" in the "skincare products" branch is 0.3. The frequency of "foundation" is 8, and the frequency of "moisturizing essence" is 10. The semantic association strength r = 0.3 * log(8 + 10 + 1) ≈ 0.3 * 1.2788 = 0.3836. Based on this, a dynamic association path of "foundation - moisturizing essence" is constructed.
[0082] Step 2044: The dynamic association path with the lowest path strength is identified as the target association path, and this target association path is added to the initial structure tree to obtain the hot word association network. Path strength is the sum of the semantic association strengths between nodes in the dynamic association path.
[0083] Furthermore, the recommendation device calculates the path strength of each dynamic association path, which is the sum of the semantic association strengths between nodes in the dynamic association path. Among all dynamic association paths, the path with the lowest path strength is selected as the target association path and added to the initial structure tree, ultimately forming a complete hot word association network.
[0084] In one embodiment, there are two dynamic association paths: path one, "foundation liquid - moisturizing essence," has a path strength of 0.3836, and path two, "base makeup product - facial cleanser," has a path strength of 0.521. Path one has the lowest path strength, so it is identified as the target association path and added to the initial structure tree, resulting in a hot word association network that includes the original hierarchical relationships and the "foundation liquid - moisturizing essence" association path.
[0085] The hot word association network constructed in this embodiment of the invention not only retains the hierarchical relationship of hot words within a single content, but also realizes the association of hot words in different content through cross-branch relationships. The determination of core branches highlights important content, and the addition of dynamic association paths enhances the relevance and dynamism of the structure tree. This makes the hot word association network comprehensively and accurately reflect the overall relationship network of hot words on the platform, providing an efficient hot word structure foundation for subsequent user interest hot word matching and personalized recommendations, thereby improving the accuracy of recommendations and user experience.
[0086] Furthermore, the content recommendation apparatus provided by the present invention will be described below. The content recommendation apparatus described below can be referred to in correspondence with the content recommendation method described above.
[0087] Optional, refer to Figure 2 , Figure 2 This is a schematic diagram of the content recommendation device provided by the present invention. The content recommendation device includes...
[0088] The data collection module 210 is used to collect real-time hot word data for each piece of content currently being published on the platform. The real-time hot word data includes the hot word text and the frequency of occurrence of each hot word text. The hot word relationship construction module 220 is used to construct a hot word association network based on the real-time hot word data of each content; each node in the hot word association network represents the hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. User profile analysis module 230 is used to perform user profile analysis based on user profile data to obtain user interest hot words; user profile data includes user's historical viewing records, historical interaction records and interest tags in registration records; The hot word association mapping module 240 is used to perform association mapping based on the hot word association network of each content and user interest hot words to obtain hot word association mapping results. The hot word association mapping results include the number of nodes in the hot word association network where user interest hot words appear, as well as the hierarchical information and frequency of each node in the hot word association network. The content recommendation module 250 is used to determine the matching degree value between each content and the user's interest hot words based on the hot word association mapping results of each content, and to generate a personalized content recommendation list based on the matching degree value of each content.
[0089] This invention constructs a hot word association network based on real-time hot word data for each content. By dividing hot words into levels and frequencies, it delves into the deep semantic relationships and features of the content. Through user profile analysis of user profile data, it accurately captures user interest hot words. Then, it uses the constructed hot word association network and user interest hot words for association mapping. Combining the number of nodes in the hot word association network for user interest hot words, as well as the level information and frequency of each node in the hot word association network, it achieves a precise association between content features and user interest features. Finally, based on the matching degree value between the content and user interest hot words determined by the hot word association mapping results, it generates a personalized content recommendation list. This ensures that the recommendation results meet the personalized needs of users and can dynamically capture content changes and user interests, improving the accuracy and diversity of recommendations and avoiding homogenization and poor timeliness.
[0090] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it performs the following steps: Collect real-time hot word data for each piece of content currently being published on the platform; real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; A hot word association network is constructed based on real-time hot word data for each content. Each node in the hot word association network represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. User profile analysis is performed based on user profile data to obtain user interest hot words; user profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records. Based on the hot word association network of each content and the hot words of user interest, an association mapping is performed to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; Based on the hot word association mapping results for each content, the matching degree value between each content and the user's interest hot words is determined, and a personalized content recommendation list is generated based on the matching degree value of each content.
[0091] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it performs the following steps: Collect real-time hot word data for each piece of content currently being published on the platform; real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; construct a hot word association network based on the real-time hot word data for each piece of content; each node in the hot word association network represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence; User profile analysis is performed based on user profile data to obtain user interest hot words; user profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records. Based on the hot word association network of each content and the hot words of user interest, an association mapping is performed to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; Based on the hot word association mapping results for each content, the matching degree value between each content and the user's interest hot words is determined, and a personalized content recommendation list is generated based on the matching degree value of each content.
[0092] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the content recommendation method provided by the above methods, the method including: Collect real-time hot word data for each piece of content currently being published on the platform; real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; A hot word association network is constructed based on real-time hot word data for each content. Each node in the hot word association network represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. User profile analysis is performed based on user profile data to obtain user interest hot words; user profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records. Based on the hot word association network of each content and the hot words of user interest, an association mapping is performed to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; Based on the hot word association mapping results for each content, the matching degree value between each content and the user's interest hot words is determined, and a personalized content recommendation list is generated based on the matching degree value of each content.
[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A content recommendation method, characterized in that, include: Collect real-time hot word data for each piece of content currently being published on the platform; The real-time hot word data includes hot word texts and the frequency of occurrence of each hot word text; A hot word association network is constructed based on real-time hot word data for each content. In the hot word association network, each node represents the hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. User profile analysis is performed based on user profile data to obtain user interest hot words; the user profile data includes user's historical viewing records, historical interaction records, and interest tags in registration records. Based on the hot word association network of each content and the hot words of user interest, an association mapping is performed to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; Based on the hot word association mapping results for each content, the matching degree value between each content and the user's interest hot words is determined, and a personalized content recommendation list is generated based on the matching degree value of each content.
2. The content recommendation method according to claim 1, characterized in that, The construction of a hot word association network based on real-time hot word data for each content includes: For any two first hot word texts and second hot word texts in the same content, if the text similarity between the first hot word text and the second hot word text is greater than or equal to the similarity threshold, then the text inclusion relationship between the first hot word text and the second hot word text is determined. Based on the text inclusion relationship, a set of node relationships is determined, and based on hierarchical clustering combined with the set of node relationships, a hot word association network with the same content is constructed in the same branch. Based on the hot word association network with different content, construct cross-branch relationships in the hot word association network within the same branch; The hot word association network is constructed based on the same branch and cross branch relationships of the hot word association network.
3. The content recommendation method according to claim 2, characterized in that, The construction of the hot word association network based on the same branch and cross-branch relationships of the hot word association network includes: Based on the cross-branch subordinate relationship in the hot word association network, a cross-content hierarchical structure is constructed. The hot word association network within the same content maintains the original hierarchical subordinate relationship within the same branch. The frequency of occurrence of each node is added as attribute information to obtain the initial structure tree. Based on the sum of the occurrence frequencies of all nodes in each branch of the initial structure tree, and combined with the level depth of each branch in the initial structure tree, the importance index of each branch is determined; the level depth of the root node is 1, and the depth increases by 1 for each level downward. Branches with an importance index greater than or equal to a preset index threshold are identified as core branches, and dynamic association paths are constructed based on the semantic similarity and semantic association strength of nodes in any two core branches. The dynamic association path with the lowest path strength is identified as the target association path, and the target association path is added to the initial structure tree to obtain the hot word association network; the path strength is the sum of the semantic association strengths between nodes in the dynamic association path.
4. The content recommendation method according to claim 2, characterized in that, The construction of cross-branch relationships in the hot word association network based on different content within the same branch includes: For any two first target content's first hot word association network branches and the second target content's second hot word association network branches, determine the semantic similarity between the first target hot word text and the second target hot word text based on the first target hot word text in the first hot word association network branch and the second target hot word text in the second hot word association network branch. Based on the semantic similarity combined with the first occurrence frequency of the first target hot word text and the second occurrence frequency of the second target hot word text, the semantic association strength between the first target hot word text and the second target hot word text is determined; The maximum semantic association strength between hot word texts is determined as the branch association strength between the same branch of the first hot word association network and the same branch of the second hot word association network. If the branch association strength is greater than or equal to the preset association strength threshold, and the root nodes of the same branch of the first hot word association network and the same branch of the second hot word association network have a semantic inclusion relationship, then it is determined that there is a cross-branch subordinate relationship between the same branch of the first hot word association network and the same branch of the second hot word association network. Based on the cross-branch subordinate relationships between the same branches of all hot word association networks, the cross-branch relationship of the hot word association network is determined.
5. The content recommendation method according to claim 4, characterized in that, The formula for calculating the semantic association strength is: r(a,b)=s(a,b)*log[f(a)+f(b)+1]; Where r(a,b) represents the semantic association strength between the first target hot word text h(a) and the second target hot word text h(b), s(a,b) represents the semantic similarity between the first target hot word text h(a) and the second target hot word text h(b), f(a) represents the first occurrence frequency of the first target hot word text h(a), f(b) represents the second occurrence frequency of the second target hot word text h(b), and log() represents the logarithmic function.
6. The content recommendation method according to claim 2, characterized in that, The step of determining the node relationship set based on the text inclusion relationship includes: If the semantic inclusion relationship is such that the first semantic range of the first hot word text completely includes the second semantic range of the second hot word text, then the first hot word text is determined to be the first parent-child node of the second hot word text; If the semantic inclusion relationship is that the second semantic range completely includes the first semantic range, then the second hot word text is determined to be the second parent-child node relationship of the first hot word text; If the semantic inclusion relationship is that the first semantic range does not completely include the second semantic range, then the first hot word text and the second hot word text are determined to be parent node compatible sibling nodes; The first parent-child node relationship, the second parent-child node relationship, and the parent node compatible sibling node relationship are determined as a set of node relationships.
7. The content recommendation method according to any one of claims 2 to 6, characterized in that, In each branch of the hot word association network, the hot word text without a parent node is used as the root node, and all direct child nodes are connected to the root node according to the parent-child node relationship in the node relationship set. For each first target child node, all direct child nodes are connected to the first target child node according to the parent-child node relationship in the node relationship set. For any two second target nodes, parent-child node connections and sibling node connections are established according to the parent-sibling node relationship in the node relationship set. Nodes in each branch of the hot word association network form a hierarchical chain according to their subordinate relationships.
8. A content recommendation device, characterized in that, Applied to the content recommendation method as described in any one of claims 1 to 7; The content recommendation device includes: The data collection module is used to collect real-time hot word data for each piece of content currently being published on the platform; the real-time hot word data includes hot word text and the frequency of occurrence of each hot word text; The hot word relationship construction module is used to build a hot word association network based on real-time hot word data for each content. Each node in the hot word association network represents a hot word text, the connection relationship between nodes represents the subordinate relationship between hot word levels, and the attribute information of the nodes represents the frequency of occurrence. The user profile analysis module is used to perform user profile analysis based on user profile data to obtain user interest hot words; the user profile data includes the user's historical viewing records, historical interaction records, and interest tags in the registration records. The hot word association mapping module is used to perform association mapping based on the hot word association network of each content and the hot words of user interest to obtain the hot word association mapping result; the hot word association mapping result includes the number of nodes in the hot word association network where the hot words of user interest appear, as well as the hierarchical information and frequency of each node in the hot word association network; The content recommendation module is used to determine the matching degree between each content and the user's interest hot words based on the hot word association mapping results of each content, and to generate a personalized content recommendation list based on the matching degree value of each content.
9. An electronic device, comprising: Memory, used to store computer software programs; A processor for reading and executing the computer software program, characterized in that, when the processor executes the computer software program, it implements the content recommendation method as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium, wherein a computer software program is stored therein, characterized in that, When the computer software program is executed by a processor, it implements the content recommendation method as described in any one of claims 1 to 7.