Information push channel recommendation method based on content attribute and audience feature fusion

Abstract

The invention relates to the technical field of information push, and discloses an information push channel recommendation method based on fusion of content attributes and audience features, and the method specifically comprises the following steps: 1, data collection: firstly extracting content attribute features, and then extracting audience features; 2, data preprocessing: constructing a time sequence behavior graph; 3, feature fusion modeling: using a double-tower neural network to code content attributes and audience features respectively, and generating fusion features through a cross attention layer; 4, channel matching model training: constructing a training set based on historical push data, and training an XGBoost or deep sorting model by taking a channel click rate as a supervision signal; according to the method, more accurate information pushing is realized by fusing the content attributes and the audience features, and information and users can be more comprehensively understood by extracting the content attribute features and the audience features and carrying out feature fusion modeling, so that the pushing accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of information push technology, and in particular to an information push channel recommendation method based on the fusion of content attributes and audience characteristics. Background Art

[0002] Information push, also known as "web broadcasting," is a new technology that uses certain technical standards or protocols to reduce information overload by regularly delivering information that users need on the internet. By automatically delivering information to users, push technology reduces the time spent searching the internet. It searches and filters information based on user interests and regularly pushes it to users, helping them efficiently discover valuable information. Technically, information push is a comprehensive field based on multiple technologies, including data mining, natural language processing, and the internet. Delivering the right information to the right people is a challenging task. This process requires thorough information analysis, detailed characterization of user interests and behaviors, and effective matching of these two. Information push has numerous applications across the industry and is a very popular technology trend on the internet today.

[0003] Existing information push technology only makes recommendations based on content keywords or user historical behavior, and does not fully integrate content attributes and audience characteristics. This results in low push channel matching accuracy, and it is difficult to update recommendation strategies in real time based on user feedback. It is unable to adapt to changes in content attributes or user interest migration. At the same time, in multi-platform scenarios, there is a lack of quantitative analysis of the synergy between channels.

[0004] To this end, we propose an information push channel recommendation method based on the fusion of content attributes and audience characteristics. Summary of the Invention

[0005] The present invention mainly solves the technical problems existing in the above-mentioned prior art and provides an information push channel recommendation method based on the fusion of content attributes and audience characteristics.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for recommending information push channels based on the integration of content attributes and audience characteristics, characterized in that it specifically includes the following steps:

[0007] Step 1: Data collection: First extract content attribute features, then extract audience features;

[0008] Step 2: Data preprocessing: constructing a time series behavior graph;

[0009] Step 3: Feature fusion modeling: Use a dual-tower neural network to encode content attributes and audience characteristics respectively, and generate fused features through a cross-attention layer;

[0010] Step 4: Channel matching model training: Build a training set based on historical push data, use channel click-through rate as a supervision signal, and train an XGBoost or deep ranking model;

[0011] Step 5: Real-time recommendation: Deploy an online inference engine to generate recommendation results based on real-time user context;

[0012] Step 6: Real-time feedback: Continuously optimize the model through A / B testing.

[0013] Preferably, extracting content attribute features in the first step specifically involves extracting text topics, sentiment analysis, and image classification through NLP tools.

[0014] Preferably, extracting audience features in the first step specifically involves extracting user device type, geographic location, and historical click behavior from the log system.

[0015] Preferably, the second step of constructing the time series behavior graph specifically includes the following steps:

[0016] S1: Build a behavior chain: Build a user behavior time series, connecting the user's behavior data at different time points to form a behavior chain;

[0017] S2: Processing behavior chains: Graphing the behavior chains, representing elements such as users, behaviors, and time in the form of nodes and edges, and constructing a time-series behavior graph;

[0018] S3: Mining user behavior patterns: In the time series behavior graph, based on the time sequence and correlation of user behaviors, user behavior patterns are mined to provide a basis for subsequent feature fusion modeling.

[0019] Preferably, the feature fusion modeling in the third step specifically includes the following steps:

[0020] S1: Receiving content attribute features and audience features: Using a dual-tower neural network to receive pre-processed content attribute features and audience features, the content attribute features are input into the content tower, and the audience features are input into the audience tower;

[0021] S2: Feature encoding: In the content tower and audience tower, the input features are encoded through the multi-layer perceptron to obtain the content attribute feature vector and audience feature vector respectively;

[0022] S3: Fusion of content attribute features and audience characteristics: The content attribute feature vector and the audience feature vector are input into the cross attention layer. By calculating the attention weight, the content attribute features and audience characteristics are fused to generate a fused feature vector.

[0023] S4: Input fully connected layer: Input the fused feature vector into the fully connected layer, and obtain the final fused feature representation through a nonlinear activation function for subsequent channel matching model training.

[0024] Preferably, the channel matching model training in the fourth step is specifically to generate a multi-channel combination recommendation strategy by quantifying the channel synergy effect.

[0025] Preferably, the real-time recommendation in the fifth step is specifically based on the real-time user context, including the current time, user location, user device information and the user's recent behavior data, combined with a pre-trained channel matching model to generate a personalized information push channel recommendation list for the user. The recommendation list will be sorted from high to low according to the predicted channel click-through rate to ensure that the channel most likely to attract the user's attention is ranked first, thereby improving the efficiency and effectiveness of information push.

[0026] Preferably, the real-time feedback in the sixth step is specifically to compare the performance of different versions of recommendation strategies in actual applications through A / B testing, collect key indicator data of user clicks and conversions, analyze the impact of different strategies on user behavior, and continuously optimize model parameters and recommendation logic based on test results to improve the accuracy of information push and user satisfaction.

[0027] The present invention provides an information push channel recommendation method based on the fusion of content attributes and audience characteristics.

[0028] It has the following beneficial effects:

[0029] 1. This information push channel recommendation method based on the fusion of content attributes and audience characteristics achieves more accurate information push by integrating content attributes and audience characteristics. By extracting content attribute features and audience characteristics and performing feature fusion modeling, it can more comprehensively understand information and users, thereby improving the accuracy of push. By constructing a time-series behavior graph and mining user behavior patterns, it achieves a detailed characterization of user behavior, clearly presents user behavior data at different time points, and mines user behavior patterns, providing strong support for subsequent feature fusion modeling. Through real-time recommendation and real-time feedback mechanisms, it achieves continuous optimization of push strategies. It can generate recommendation results based on real-time user context, and collect key user click and conversion indicator data through A / B testing. It continuously optimizes model parameters and recommendation logic to ensure that push strategies are always consistent with user interests. By quantifying channel synergy, it further improves the effectiveness of multi-channel combination recommendation strategies, enables a more scientific combination of multiple channels for push, thereby improving the coverage rate and user satisfaction of information push. It can also dynamically adjust the weight of different channels in the push strategy based on channel characteristics and user preferences, achieving more refined push management.

[0030] 2. This information push channel recommendation method based on the fusion of content attributes and audience characteristics has significantly improved the efficiency of information push and user satisfaction through innovative feature fusion technology and real-time optimization mechanism. It not only takes into account content attributes, but also deeply analyzes audience characteristics, realizing the effective combination of the two in the push process. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is the overall flow chart of the method of the present invention;

[0032] Figure 2 This is a flow chart of the method for constructing a time sequence behavior graph of the present invention;

[0033] Figure 3 This is a flow chart of the feature fusion modeling method of the present invention. DETAILED DESCRIPTION

[0034] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely illustrative, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0035] The structures, proportions, sizes, etc. illustrated in this specification are intended only to complement the contents disclosed herein and to facilitate understanding and reading by persons familiar with the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall still fall within the scope of the technical contents disclosed herein.

[0036] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0037] In the description of the embodiments of the present invention, it should be noted that the terms "center," "upper," "lower," "inner," "outer," and "side" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or the orientations or positional relationships in which the inventive product is typically placed when in use. These terms are intended solely to facilitate the description of the present invention and to simplify the description, and are not intended to indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and the like are used solely to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0038] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the embodiments of the present invention according to specific circumstances.

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] Example 1: Information push channel recommendation method based on the fusion of content attributes and audience characteristics, such as Figure 1 As shown, it is characterized in that it specifically includes the following steps:

[0041] Step 1: Data collection: First extract content attribute features, then extract audience features;

[0042] Step 2: Data preprocessing: constructing a time series behavior graph;

[0043] Step 3: Feature fusion modeling: Use a dual-tower neural network to encode content attributes and audience characteristics respectively, and generate fused features through a cross-attention layer;

[0044] Step 4: Channel matching model training: Build a training set based on historical push data, use channel click-through rate as a supervision signal, and train an XGBoost or deep ranking model;

[0045] Step 5: Real-time recommendation: Deploy an online inference engine to generate recommendation results based on real-time user context;

[0046] Step 6: Real-time feedback: Continuously optimize the model through A / B testing.

[0047] Example 2: Based on Example 1, Figure 1As shown in the figure, the first step of extracting content attribute features involves extracting text topics, sentiment analysis, and image classification through NLP tools. The first step of extracting audience features involves extracting user device type, geographic location, and historical click behavior from the log system. By integrating content attributes and audience characteristics, more accurate information push is achieved. By extracting content attribute features and audience characteristics and performing feature fusion modeling, a more comprehensive understanding of information and users is achieved, thereby improving push accuracy. By constructing a time-series behavior graph and mining user behavior patterns, a detailed description of user behavior is achieved. The user's behavior data at different time points can be clearly displayed, and the user's behavior patterns can be mined, providing strong support for subsequent feature fusion modeling. Through real-time recommendation and real-time feedback mechanisms, push strategies are continuously optimized. Recommendations can be generated based on real-time user context. Through A / B testing, key user click and conversion metrics are collected, and model parameters and recommendation logic are continuously optimized to ensure that push strategies are always consistent with user interests. By quantifying channel synergy, the effectiveness of multi-channel combination recommendation strategies is further improved. Multiple channels can be more scientifically combined for push, thereby improving information push coverage and user satisfaction. The weight of different channels in the push strategy can also be dynamically adjusted based on channel characteristics and user preferences, achieving more refined push management.

[0048] Example 3: Based on Example 1 and Example 2, Figure 2 As shown in the figure, the second step of constructing the timing behavior graph specifically includes the following steps:

[0049] S1: Build a behavior chain: Build a user behavior time series, connecting the user's behavior data at different time points to form a behavior chain;

[0050] S2: Processing behavior chains: Graphing the behavior chains, representing elements such as users, behaviors, and time in the form of nodes and edges, and constructing a time-series behavior graph;

[0051] S3: Mining user behavior patterns: In the time series behavior graph, based on the time sequence and correlation of user behaviors, user behavior patterns are mined to provide a basis for subsequent feature fusion modeling.

[0052] Example 4: Based on Example 1, Example 2 and Example 3, Figure 3 As shown, the feature fusion modeling in the third step specifically includes the following steps:

[0053] S1: Receiving content attribute features and audience features: Using a dual-tower neural network to receive pre-processed content attribute features and audience features, the content attribute features are input into the content tower, and the audience features are input into the audience tower;

[0054] S2: Feature encoding: In the content tower and audience tower, the input features are encoded through the multi-layer perceptron to obtain the content attribute feature vector and audience feature vector respectively;

[0055] S3: Fusion of content attribute features and audience characteristics: The content attribute feature vector and the audience feature vector are input into the cross attention layer. By calculating the attention weight, the content attribute features and audience characteristics are fused to generate a fused feature vector.

[0056] S4: Input fully connected layer: Input the fused feature vector into the fully connected layer, and obtain the final fused feature representation through a nonlinear activation function for subsequent channel matching model training.

[0057] Example 5: Based on Example 1, Example 2, Example 3 and Example 4, Figure 1 As shown, in the fourth step, channel matching model training involves quantifying channel synergies and generating a multi-channel combination recommendation strategy. In the fifth step, real-time recommendations are generated based on real-time user context, including the current time, user location, device information, and recent user behavior data, combined with the pre-trained channel matching model to generate a personalized list of recommended channels for information push. This list is sorted from high to low by predicted channel click-through rates to ensure that the channels most likely to attract the user's attention are prioritized, thereby improving the efficiency and effectiveness of information push. In the sixth step, real-time feedback is provided by A / B testing, comparing the performance of different recommendation strategies in real-world applications. Key user click and conversion metrics are collected, and the impact of different strategies on user behavior is analyzed. Based on the test results, model parameters and recommendation logic are continuously optimized to improve information push accuracy and user satisfaction. Through innovative feature fusion technology and real-time optimization mechanisms, the efficiency and user satisfaction of information push are significantly improved. This not only considers content attributes but also deeply analyzes audience characteristics, effectively integrating the two in the push process.

[0058] The working principle of the present invention is as follows: First, a rich set of content attributes and audience characteristics information is acquired through the data collection step. In the data preprocessing stage, the constructed temporal behavior graph can accurately capture the temporal relationships and patterns of user behavior, providing a solid foundation for subsequent feature fusion. In the feature fusion modeling part, a dual-tower neural network is used to process content attributes and audience characteristics respectively, and a cross-attention mechanism is used to achieve deep fusion of features. This step effectively improves the accuracy and richness of feature representation. In the channel matching model training stage, by introducing historical push data as a training set and combining channel click-through rate as a supervision signal, an efficient and adaptable matching model can be trained. The real-time recommendation link quickly generates a personalized information push channel recommendation list based on the user's real-time context information, ensuring the timeliness and pertinence of information push. Finally, through a real-time feedback mechanism, the model parameters and recommendation logic are continuously optimized, allowing the entire push system to continuously evolve and better adapt to changes in user needs and interests. By fully integrating content attributes and audience characteristics, high-precision matching and personalized recommendations of information push channels are achieved, effectively improving the efficiency of information push and user satisfaction.

[0059] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An information push channel recommendation method based on the integration of content attributes and audience characteristics, characterized in that: The specific steps include: Step 1: Data collection: First extract content attribute features, then extract audience features; Step 2: Data preprocessing: constructing a time series behavior graph; Step 3: Feature fusion modeling: Use a dual-tower neural network to encode content attributes and audience characteristics respectively, and generate fused features through a cross-attention layer; Step 4: Channel matching model training: Build a training set based on historical push data, use channel click-through rate as a supervision signal, and train an XGBoost or deep ranking model; Step 5: Real-time recommendation: Deploy an online inference engine to generate recommendation results based on real-time user context; Step 6: Real-time feedback: Continuously optimize the model through A / B testing.

2. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The first step of extracting content attribute features specifically involves extracting text topics, sentiment analysis, and image classification through NLP tools.

3. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The first step of extracting audience features specifically involves extracting user device type, geographic location, and historical click behavior from the log system.

4. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The second step of constructing the timing behavior graph specifically includes the following steps: S1: Build a behavior chain: Build a user behavior time series, connecting the user's behavior data at different time points to form a behavior chain; S2: Processing behavior chains: Graphing the behavior chains, representing elements such as users, behaviors, and time in the form of nodes and edges, and constructing a time-series behavior graph; S3: Mining user behavior patterns: In the time series behavior graph, based on the time sequence and correlation of user behaviors, user behavior patterns are mined to provide a basis for subsequent feature fusion modeling.

5. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The feature fusion modeling in the third step specifically includes the following steps: S1: Receiving content attribute features and audience features: Using a dual-tower neural network to receive pre-processed content attribute features and audience features, the content attribute features are input into the content tower, and the audience features are input into the audience tower; S2: Feature encoding: In the content tower and audience tower, the input features are encoded through the multi-layer perceptron to obtain the content attribute feature vector and audience feature vector respectively; S3: Fusion of content attribute features and audience characteristics: The content attribute feature vector and the audience feature vector are input into the cross attention layer. By calculating the attention weight, the content attribute features and audience characteristics are fused to generate a fused feature vector. S4: Input fully connected layer: Input the fused feature vector into the fully connected layer, and obtain the final fused feature representation through a nonlinear activation function for subsequent channel matching model training.

6. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The channel matching model training in the fourth step is specifically to generate a multi-channel combination recommendation strategy by quantifying the channel synergy effect.

7. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The real-time recommendation in the fifth step is specifically to generate a personalized information push channel recommendation list for the user based on the real-time user context, including the current time, user location, user device information and the user's recent behavior data, combined with the pre-trained channel matching model. The recommendation list will be sorted from high to low according to the predicted channel click-through rate to ensure that the channel most likely to attract the user's attention is ranked first, thereby improving the efficiency and effectiveness of information push.

8. The information push channel recommendation method based on the fusion of content attributes and audience characteristics according to claim 1 is characterized by: The real-time feedback in the sixth step is specifically to compare the performance of different versions of recommendation strategies in actual applications through A / B testing, collect key indicator data of user clicks and conversions, analyze the impact of different strategies on user behavior, and continuously optimize model parameters and recommendation logic based on test results to improve the accuracy of information push and user satisfaction.

Images