Advertisement putting position management system based on preference degree analysis

Abstract

The invention relates to the technical field of advertisement putting position management, in particular to an advertisement putting position management system based on preference degree analysis. The system comprises a positioning and predicting unit, an analyzing and eliminating unit and an evaluating and putting unit. According to the invention, the evaluation and delivery unit evaluates the emotion of the user according to the collected facial expression image data and voice intonation data of the user and the captured facial expression image data of the user, the evaluated emotion of the user is'pleasure 'and'anger', and the advertisement delivery is carried out according to the evaluated emotion of the user and the collected surrounding environment information of the position point. The advertisement effect can be more accurately improved by evaluating the emotion of the user, the emotion change of the user before the advertisement is pushed is known, the advertisement type can be adjusted according to the emotion change of the user so as to improve the attraction of the advertisement, and meanwhile more attractive advertisement content can be customized according to the emotion of the user and the surrounding environment information. And the click rate and the conversion rate of the advertisement are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of advertisement placement location management, and in particular to an advertisement placement location management system based on preference analysis. Background Art

[0002] Analyzing the preference of advertising placement can help advertisers more effectively place their ads in the target audience group's location, which is most attractive to the specific audience group. The benefits of analyzing the preference of advertising placement can improve the effect of advertising placement, increase user participation in advertising, and also provide advertisers with more accurate advertising placement strategies. When a user stays at a certain place, advertisers will only place ads based on the information around the user's place of stay, but will not place ads based on the user's emotional changes, thereby reducing the user's experience. Since the user likes milk tea, the user has a high preference for advertisements about milk tea, so the user will The user moves towards a milk tea shop, but during the movement he browses about unhygienic issues about milk tea, which causes the user's emotions to become unstable. When the user stays around the milk tea shop, since the advertiser does not mine the user's web browsing history during the movement, a large number of milk tea-related ads are served to the user, causing the user's emotions to change from unstable to angry, reducing the user experience. At the same time, the user closes or ignores the ad content, reducing the click-through rate and conversion rate of the ad. In order to avoid a decrease in the click-through rate and conversion rate of the ad due to changes in the user's emotions, we provide an ad placement location management system based on preference analysis. Summary of the invention

[0003] The purpose of the present invention is to provide an advertisement placement location management system based on preference analysis to solve the problems raised in the above background technology.

[0004] To achieve the above-mentioned purpose, the present invention provides an advertisement placement location management system based on preference analysis, comprising a positioning and prediction unit, an analysis and exclusion unit, and an evaluation and placement unit;

[0005] The positioning and prediction unit obtains the user's favorite advertisement data, user movement pattern and historical data and performs positioning, collects the user's facial expression image data and voice intonation data according to the positioning information, and then predicts the user's stay location based on the acquired historical data;

[0006] The analysis and exclusion unit is used to receive the user's favorite advertisements and the predicted user's stay location points obtained by the positioning and prediction unit, and perform a preference analysis of the favorite advertisements and identify low preference factors based on the obtained user's favorite advertisements, and then collect surrounding environment information based on the predicted user's stay location points, and then exclude low preference factor data from the collected surrounding environment information;

[0007] The evaluation and delivery unit is used to receive the user facial expression image data, voice and intonation data collected by the positioning and prediction unit and the location point surrounding environment information collected by the analysis and exclusion unit, and evaluate the user's emotions based on the collected user facial expression image data and voice and intonation data, and then deliver advertisements based on the evaluated user emotions and the collected location point surrounding environment information.

[0008] As a further improvement of the technical solution, the positioning and prediction unit includes a positioning acquisition module and a position prediction module;

[0009] The positioning acquisition module is used to obtain the user's favorite advertising data, user movement patterns and historical data, obtain the user's location data in real time according to the obtained user movement patterns for positioning, and use surrounding cameras to collect the user's facial expression image data and voice intonation data according to the positioning information;

[0010] The location prediction module is used to receive the positioning information, the acquired user movement pattern and the acquired historical data in the positioning acquisition module, obtain the movement trajectory data from the user movement pattern data, and analyze the user's activity range according to the movement trajectory, and then use the random model to predict the user's stay location point according to the analyzed user activity range and the acquired historical data, and locate according to the predicted passing location point and use the camera around the predicted passing location point to capture the user's facial expression image data.

[0011] As a further improvement of the technical solution, the analysis exclusion unit includes an analysis comparison module and a data exclusion module;

[0012] The analysis and comparison module is used to receive the user's favorite advertisements obtained in the positioning collection module, and use association rule mining to perform a preference analysis of the favorite advertisements based on the obtained user's favorite advertisements, and then use the analyzed preference of the favorite advertisements to compare with the set preference of the advertisements;

[0013] The data exclusion module is used to receive the user's stay location predicted in the location prediction module and the low preference factor data in the analysis and comparison module, perform positioning according to the predicted user's stay location, and then collect surrounding environment information based on the positioning information and using a camera, and then exclude the low preference factor data from the collected surrounding environment information to obtain the excluded environment information.

[0014] As a further improvement of the technical solution, the specific comparison conditions in the analysis and comparison module include:

[0015] Case ①: When the degree of preference of the analyzed favorite advertisement is greater than the set degree of preference of the advertisement, it indicates that the obtained user favorite advertisement has a high degree of preference, and a preference model is established based on the obtained user favorite advertisement. The preference model is used to collect the obtained user favorite advertisements, and the degree of preference of the obtained user favorite advertisements is higher than the set degree of preference of the advertisement. The preference model is used to classify the collected user favorite advertisements into advertising types. The types of favorite advertisements in the preference model are divided into humorous, healing, warm and cheerful.

[0016] Case ②: When the preference of the analyzed preferred advertisement is less than the set advertisement preference, it indicates that the obtained user preferred advertisement has a low preference degree, and the low preference factor is identified based on the obtained user preferred advertisement.

[0017] As a further improvement of the present technical solution, the evaluation and delivery unit is used to receive the user facial expression image data, voice and intonation data collected in the positioning acquisition module, the user facial expression image data captured in the location prediction module, and the predicted user stay location point, receive the preference model in the analysis and comparison module and the environmental information excluded in the data exclusion module, and the collected location point surrounding environmental information, and use a support vector machine to evaluate the user's emotions based on the collected user facial expression image data, voice and intonation data and the captured user facial expression image data, the evaluated user emotions are "pleasure" and "anger", respectively, and delivery of advertisements is performed based on the evaluated user emotions and the collected location point surrounding environmental information.

[0018] As a further improvement of the technical solution, the specific delivery steps in the evaluation delivery unit include:

[0019] S1. When the user's emotion is evaluated as positive and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertisement types in the preference model, and the preferred advertisement types related to the excluded environmental information are matched, and then the preferred advertisement types are sorted according to the preference degree of the related preferred advertisement types;

[0020] S2. When the evaluated user emotion is anger, and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertising types in the preference model to match the preferred advertising types related to the excluded environmental information, and then a recurrent neural network is used to analyze the cause of anger based on the user's web browsing records within the analyzed activity range time, and the related preferred advertising types containing the analyzed cause of anger are excluded based on the analyzed cause of anger, and the related preferred advertising types excluding the cause of anger are obtained, and then humorous type advertisements are extracted from the related preferred advertising types excluding the cause of anger, and then the extracted humorous type advertisements are compared and ranked using the preference degree of the extracted humorous type advertisements.

[0021] As a further improvement of the technical solution, the specific ranking of S1 in the evaluation and delivery unit includes:

[0022] Case 1: If the ranked ads are ranked from low to high in terms of their likeability, the most likeable ad is selected and placed;

[0023] Case 2: If the ranked advertisement preferences are from low to high, and there are two or more related preference advertisements with the highest advertisement preference, then the related preference advertisement with the highest advertisement preference is randomly selected for advertisement delivery.

[0024] As a further improvement of the technical solution, the specific ranking of S2 in the evaluation delivery unit includes:

[0025] Case 1: If the ranked advertisements are ranked from low to high in terms of their likeability, the humorous advertisement with the highest likeability is selected and placed;

[0026] Case 2: If the ranked advertisement preference is from low to high, and there are two or more humorous type advertisements with the highest preference, the extracted humorous type advertisements are evaluated for humorousness using random forest, and the humorous type advertisement with the highest humorousness is selected for advertisement delivery.

[0027] As a further improvement of the technical solution, the evaluation and delivery unit S2 uses random forest to evaluate the humor level of humorous advertisements to implement the following steps:

[0028] Step 1: First, collect the extracted humorous advertisements and use them as training humorous data sets. Each humorous advertisement text is used as a humorous sample, and its humorous degree score is used as the corresponding label. Then, feature extraction is performed on the humorous samples. The word embedding model can be used to convert the text into a vector representation, and the feature word frequency and text length are extracted.

[0029] Step 2: Use the training humor data set to train the random forest regression model. The random forest constructs multiple decision trees and obtains the number of decision trees N in the random forest and the prediction result f of the decision tree. i , f i is the prediction result of the i-th decision tree;

[0030] Step 3: Assume that we have a humor training set (D = {(x1, y1), (x2, y2), ..., (x n ,y n )}, where x i is a characteristic of humorous advertising texts, i is the corresponding humor score, and then the prediction results of the decision tree are summed to obtain the sum of the decision tree prediction resultsf =f1+f2+...+f n ;

[0031] Step 4: Use the weighted decision tree to predict the result sum f , the number of decision trees N in the random forest and the prediction results f of the decision trees i Evaluate the humor level and get the humor level PG y , the specific algorithm formula is:

[0032]

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. In the advertising placement management system of the preference analysis, the evaluation and placement unit evaluates the user's emotions based on the collected user facial expression image data, voice intonation data and captured user facial expression image data. The evaluated user emotions are "pleasure" and "anger", and advertisements are placed based on the evaluated user emotions and the collected location point surrounding environment information. Evaluating user emotions can more accurately improve advertising effects and understand the user's emotional changes before pushing advertisements. The advertisement type can be adjusted according to the user's emotional changes to enhance the attractiveness of the advertisement. At the same time, more attractive advertising content can be customized based on the user's emotions and surrounding environment information to improve the click-through rate and conversion rate of the advertisement.

[0035] 2. In the advertising placement management system of the preference analysis, the analysis and comparison module is used to perform preference analysis on the user preference advertisements obtained from the positioning collection module, and then compare the preference of the analyzed preference advertisements with the set preference of the advertisements. Providing advertising content that meets the user's preferences can enhance the user's advertising viewing experience. Users are more inclined to interact with the content that interests them. Therefore, providing advertisements that meet their preferences can reduce users' sense of interference and boredom, and enhance advertising acceptance and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is an overall block diagram of the present invention;

[0037] Figure 2 It is a block diagram of the positioning and prediction unit of the present invention;

[0038] Figure 3 The unit block diagram is excluded for the analysis of the present invention.

[0039] The meaning of each number in the figure is:

[0040] 1. Positioning and prediction unit; 11. Positioning acquisition module; 12. Position prediction module;

[0041] 2. Analysis and exclusion unit; 21. Analysis and comparison module; 22. Data exclusion module; 3. Evaluation and delivery unit. DETAILED DESCRIPTION

[0042] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0043] Example 1

[0044] The present invention provides an advertising placement management system based on preference analysis, please refer to Figure 1-Figure 3 , including a positioning and prediction unit 1, an analysis and exclusion unit 2, and an evaluation and delivery unit 3;

[0045] The positioning and prediction unit 1 obtains the user's favorite advertisement data, user movement pattern and historical data and performs positioning, and collects the user's facial expression image data and voice and intonation data based on the positioning information, and then predicts the user's stay location based on the acquired historical data; the analysis and exclusion unit 2 is used to receive the user's favorite advertisements and the predicted user's stay location obtained in the positioning and prediction unit 1, and perform a preference analysis of the favorite advertisements and identify low preference factors based on the obtained user's favorite advertisements, and then collect surrounding environment information based on the predicted user's stay location, and then exclude low preference factor data from the collected surrounding environment information; the evaluation and delivery unit 3 is used to receive the user's facial expression image data, voice and intonation data collected in the positioning and prediction unit 1 and the location point surrounding environment information collected in the analysis and exclusion unit 2, and evaluate the user's emotions based on the collected user's facial expression image data and voice and intonation data, and then deliver advertisements based on the evaluated user emotions and the collected location point surrounding environment information.

[0046] The following is a refinement of the above units, see Figure 1-Figure 3 ;

[0047] The positioning and prediction unit 1 includes a positioning acquisition module 11 and a position prediction module 12;

[0048] The positioning acquisition module 11 is used to obtain the user's favorite advertising data, user movement patterns and historical data, obtain the user's location data in real time according to the obtained user movement patterns for positioning, and use the surrounding cameras to collect the user's facial expression image data and voice intonation data according to the positioning information;

[0049] User movement patterns include user web browsing records and movement trajectories; user web browsing records include browsed websites, browsing time, and web page clicked links;

[0050] The historical data obtained includes the time spent and the data of frequently visited places;

[0051] The location prediction module 12 is used to receive the positioning information, the acquired user movement pattern and the acquired historical data in the positioning acquisition module 11, obtain the movement trajectory data from the user movement pattern data, and analyze the user's activity range according to the movement trajectory, and then use a random model (the random model belongs to the prior art) to predict the user's stay location point according to the analyzed user activity range and the acquired historical data, and locate according to the predicted passing location point and use the camera around the predicted passing location point to capture the user's facial expression image data. By predicting the user's stay location, advertisers can choose to place advertisements in areas where users are active frequently, thereby improving the exposure rate and click-through rate of advertisements and increasing the conversion effect of advertisements;

[0052] The implementation steps of using random models to predict the user's stay location are as follows:

[0053] Step ①: First collect the analyzed user activity range FW and the acquired historical data LS, and extract the activity frequency feature vector PL from the analyzed user activity range FW xl , and then extract the time stay feature vector SJ from the acquired historical data LS xl and the feature vector DD of frequently visited places xl ;

[0054] Step 2: The dynamic frequency feature vector PL xl , time dwell feature vector SJ xl , and the feature vector DD of frequently visited places xl Enter the random model SJ mx And obtain the corresponding random model parameters a0, a1, a2, a3, a0 refers to the constant term in the random model;

[0055] Step 3: Use the dynamic frequency eigenvector PL xl , time dwell feature vector SJ xl , and the feature vector DD of frequently visited places xl The user's stay location is predicted with the random model parameters a0, a1, a2, and a3, and the predicted user stay location YC is obtained. wzd , the specific algorithm formula is:

[0056] YC wzd =a0+(a1*PL xl )+(a2*SJ xl)+(a3*DD xl );

[0057] This formula is used to predict the user's stay location. By predicting the user's stay location, advertisers can more accurately target advertising audiences. By placing ads near places where users often appear, the effectiveness and conversion rate of advertising reach can be increased.

[0058] The analysis and exclusion unit 2 includes an analysis and comparison module 21 and a data exclusion module 22;

[0059] The analysis and comparison module 21 is used to receive the user's favorite advertisements obtained by the positioning collection module 11, and use association rule mining to perform a preference analysis of the favorite advertisements based on the obtained user's favorite advertisements, and then use the analyzed preference of the favorite advertisements to compare with the set preference of the advertisements. The analysis of the user's favorite advertisements can enable advertisers to place advertisements more accurately. By understanding the interests and preferences of users, advertisers can select target audiences in a targeted manner and place advertising resources in the areas that are most likely to arouse user interest, thereby improving advertising effects and conversion rates;

[0060] The implementation steps of using association rules to mine and analyze the preference of favorite advertisements are as follows:

[0061] Step 1: First collect the user’s favorite advertisements XH gg And integrate them into a data set, use the Apriori algorithm to mine frequent item sets PF from users' advertising preference data xj and association rules GL gz , specific implementation steps:

[0062] Step 1: Get the user's favorite advertisement XH gg Represented as an item set X j , where each item X represents an advertisement, and then initialize the frequent item set PF xj The size is 1, by scanning the entire data set and calculating the support of an advertisement ZC d , and then discard the support ZC d Below the set support threshold SZC d One advertisement, and the remaining advertisements form a frequent item set PF xj A collection of;

[0063] Step 2: Based on frequent item sets PF xj The set of candidate items HX is generated by using the prior properties xj , for a frequent itemset PF of size k xj , generate a candidate item set HX of size k+1 xj , ensuring that any two candidate item sets HX xj The first k-1 items of are the same;

[0064] Step 3: For the generated candidate item set HX xj , check that its subset belongs to the frequent item set PF xj If the candidate set HX xj If the subset is not frequent, then the candidate item set HX xj Delete, keep the remaining itemsets X j As frequent itemsets PF xj ;

[0065] Step 4: Repeat steps 2 and 3 until no more candidate item sets HX can be generated. xj So far, for frequent item sets PF xj Generate its non-empty subset, according to the confidence ZX of the subset d Filter out confidence ZX d Association rules that are above the set confidence threshold GL gz ;

[0066] Step 2: According to the association rule GL gz Support ZC d Indicators are used to calculate the user's preference for ads. The implementation steps are as follows:

[0067] Step 1: Get the data set in step 1 and traverse the total number of transaction records in the data set. jy , check whether it contains advertisements A and B, add the transaction records that contain A and B, and get the number of transaction records of A and B AB jy ;

[0068] Step 2: Use the total number of transaction records sum jy and the number of transaction records AB jy Calculate the support and obtain the calculated expenditure, the calculated expenditure JS zcd That is to analyze the preference of the favorite advertisements. The specific calculation formula is:

[0069] JS zcd =sum jy / AB jy ;

[0070] The specific comparison situations in the analysis and comparison module 21 include:

[0071] Case ①: When the degree of preference of the analyzed favorite advertisement is greater than the set degree of preference of the advertisement, it indicates that the obtained user favorite advertisement has a high degree of preference, and a preference model is established based on the obtained user favorite advertisement. The preference model is used to collect the obtained user favorite advertisements, and the degree of preference of the obtained user favorite advertisements is higher than the set degree of preference of the advertisement. The preference model is used to classify the collected user favorite advertisements into advertising types. The types of favorite advertisements in the preference model are divided into humorous, healing, warm and cheerful.

[0072] Case ②: When the preference of the analyzed preferred advertisement is less than the set advertisement preference, it indicates that the obtained user preferred advertisement has a low preference degree, and the low preference factor is identified based on the obtained user preferred advertisement.

[0073] The data exclusion module 22 is used to receive the user's stay location predicted by the location prediction module 12 and the low preference factor data in the analysis and comparison module 21, perform positioning according to the predicted user's stay location, collect surrounding environment information using a camera based on the positioning information, exclude the low preference factor data from the collected surrounding environment information, and obtain the excluded environment information;

[0074] When the predicted user stop location is in front of the gym, the cameras around the gym are used to obtain the environment information around the gym. The obtained environment information around the gym includes movies, food, people, milk tea, and e-sports information. Since the user liked to drink milk tea before, the user's preference for milk tea advertisements was high enough. However, due to the recent news about the hygiene of milk tea, the user's preference for milk tea-related advertisements has greatly decreased. At this time, the push of milk tea-related advertisements does not match the user's current interests, causing the user to feel dissatisfied. The user will feel that the advertisement is irrelevant or disturbing, and may choose to ignore or close the advertisement, reducing the click-through rate and conversion rate of the advertisement. In order to avoid the low preference factor data (referring to milk tea) causing the decrease in the click-through rate and conversion rate of the advertisement, the low preference factor data is excluded in advance when pushing the advertisement;

[0075] The evaluation and delivery unit 3 is used to receive the user facial expression image data, voice and intonation data collected by the positioning acquisition module 11, the user facial expression image data captured by the position prediction module 12, and the predicted user stay location point, receive the preference model in the analysis and comparison module 21 and the environmental information excluded by the data exclusion module 22, and the surrounding environmental information of the collected location point, and use the support vector machine to evaluate the user's emotions based on the collected user facial expression image data, voice and intonation data and the captured user facial expression image data, the evaluated user emotions are "pleasure" and "anger", and the advertisement is delivered according to the evaluated user emotions and the collected surrounding environmental information of the location point;

[0076] Steps to evaluate user emotions using support vector machines:

[0077] Step ①, first collect the collected user facial expression image data, voice intonation data and captured user facial expression image data and mark the user's emotions, then integrate the marked user emotion data into an emotion data set, and divide the emotion data set into a marked training data set and a test data set;

[0078] Step ②, observe that the labeled training data set contains sample data of the emotion category, and the samples have been correctly labeled, and then adjust the parameters according to the selected support vector machine model, and then use the labeled training data set to train the support vector machine model after adjustment;

[0079] Step 3: Get a test data set, which contains samples of emotion categories. Use the trained support vector machine model to predict the test data set. For samples of emotion categories, the model will output a prediction result of the emotion category.

[0080] The specific delivery steps in the evaluation delivery unit 3 include:

[0081] S1. When the user's emotion is evaluated as positive and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertisement types in the preference model, and the preferred advertisement types related to the excluded environmental information are matched, and then the preferred advertisement types are sorted according to the preference degree of the related preferred advertisement types;

[0082] When the user's emotion is evaluated as positive, and the user is about to arrive at the gym location in a few minutes, the environmental information excluding milk tea (including movie, food, person, and e-sports information after excluding milk tea) is matched with the advertising types in the preference model to match the relevant types of advertisements including movie, food, person, and e-sports, and then the preference of the relevant types of advertisements of movie A (including A1 and A2), food B (including B1 and B2), person C (including C1 and C2), and e-sports D (including D1 and D2) is ranked;

[0083] The specific sorting includes:

[0084] Case 1: If the ranked ads are ranked from low to high in terms of their likeability, the most likeable ad is selected and placed;

[0085] When the ranking of advertisement preference is movie A1<food B1<character C1<esports D1, the advertisement related to esports D1 with the highest preference is selected for delivery;

[0086] Case 2: If the ranked ads are ranked from low to high in terms of preference, and there are two or more related ads with the highest preference, then the related ad with the highest preference is randomly selected for advertising.

[0087] When the ranking preference is movie A1 < food B1 < person C1 = e-sports D1, it means that the preference of the advertisements of person C1 and e-sports D1 are both the highest, and the preference advertisements related to person C1 and e-sports D1 are randomly selected for delivery;

[0088] S2. When the user emotion is evaluated as anger and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertisement types in the preference model to match the preferred advertisement types related to the excluded environmental information, and then the recurrent neural network is used to analyze the anger cause according to the user's web browsing record within the analyzed activity range time, and the related preferred advertisement types containing the analyzed anger cause are excluded from the related preferred advertisement types according to the analyzed anger cause, and the related preferred advertisement types excluding the anger cause are obtained, and then humorous advertisements are extracted from the related preferred advertisement types excluding the anger cause, and then the extracted humorous advertisements are compared and ranked using the preference degree of the extracted humorous advertisements;

[0089] When the user's emotion is evaluated as angry, and the user will arrive at the gym location in a few minutes, the environmental information excluding milk tea related information (including movie, food, character, and e-sports information after excluding milk tea) is matched with the advertising type in the preference model to match the relevant types of advertisements including movie, food, character, and e-sports, and the user's web browsing records within the analysis activity range are mined to analyze the cause of anger. If the analyzed cause of anger is bad news about a certain person, which leads to the user's emotional anger, in order to stabilize the user's emotion and exclude related character type advertisements, the relevant preferred advertising types excluding the cause of anger are obtained (the relevant preferred advertising types include movie, food, and e-sports advertising types), and then humorous types (humorous types include movie humor, food humor, and e-sports humor) are extracted from the relevant preferred advertising types (the relevant preferred advertising types include movie, food, and e-sports advertising types), and the movie humor A is analyzed. you (Including A you1 , A you2 ), Food Humor B you (including B you1 , B you2 ), E-sports humor you (including D you1 , D you2 ) Advertisement preference ranking;

[0090] The implementation steps of using recurrent neural network to analyze the causes of anger are:

[0091] Step 1: First, collect and analyze the user's web browsing history, extract features from the user's web browsing history, extract the features of web page content, access frequency, browsing time, and click behavior, and convert the extracted features into the input data format of the RNN network;

[0092] Step 2: Mark each web browsing record with emotion, and mark the emotion into categories such as joy and anger, and use it as the label information of the RNN. Then, serialize the user's web browsing record and convert it into a sequence data format suitable for RNN input;

[0093] Step 3: Select a suitable recurrent neural network structure to capture the long-term dependencies in the sequence and build a recurrent neural network model. Take the webpage content sequence of word embedding as the input sequence, process it through the recurrent neural network layer, and output the final reason for emotional anger.

[0094] The specific sorting includes:

[0095] Case 1: If the ranked advertisements are ranked from low to high in terms of their likeability, the humorous advertisement with the highest likeability is selected and placed;

[0096] When the ranking of advertising preference is movie humor A you1 <Food Humor B you1 <Esports Humor D you1 , explaining eSports humor you1 The type of advertisement has the highest preference, and chooses e-sports humor D you1 To place advertisements;

[0097] Case 2: If the ranked advertisements are ranked from low to high in terms of preference, and there are two or more humorous advertisements with the highest preference, the extracted humorous advertisements are evaluated for humorousness using random forest, and the humorous advertisement with the highest humorousness is selected for advertisement delivery;

[0098] When the ranking of advertising preference is movie humor A you1 <Food Humor B you1 =Esports Humor D you1 , explaining food humor B you1 , E-sports humor you1 Type of advertisements has the highest preference, followed by food humor B you1 , E-sports humor you1 The humor of the advertisement is analyzed, and the food humor B is analyzed. you1 Humor level is 7, eSports humor is D you1 Humor level is 10, due to food humor B you1 Humor level is lower than e-sports humor D you1 Humor level, and select eSports Humor D you To place advertisements;

[0099] The steps to use random forest to evaluate the humor level of humorous advertisements are as follows:

[0100] Step 1: First, collect the extracted humorous advertisements and use them as training humorous data sets. Each humorous advertisement text is used as a humorous sample, and its humorous degree score is used as the corresponding label. Then, feature extraction is performed on the humorous samples. The word embedding model can be used to convert the text into a vector representation, and the feature word frequency and text length are extracted.

[0101] Step 2: Use the training humor data set to train the random forest regression model. The random forest constructs multiple decision trees and obtains the number of decision trees N in the random forest and the prediction result f of the decision tree. i , f i is the prediction result of the i-th decision tree;

[0102] Step 3: Assume that we have a humor training set (D = {(x1, y1), (x2, y2), ..., (x n ,y n )}, where x i is a characteristic of humorous advertising texts, i is the corresponding humor score, and then the prediction results of the decision tree are summed to obtain the sum of the decision tree prediction results f =f1+f2+...+f n ;

[0103] Step 4: Use the weighted decision tree to predict the result sum f , the number of decision trees N in the random forest and the prediction results f of the decision trees i Evaluate the humor level and get the humor level PG y , the specific algorithm formula is:

[0104]

[0105] This formula is used to evaluate the humor level of the extracted humorous advertisements. Humor can enable users to establish emotional resonance and establish a more positive emotional connection between advertisements and users. Evaluating the humor level of advertisements can reduce the user's angry emotional state at this time, thereby better touching the hearts of the audience.

[0106] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An advertising placement management system based on preference analysis, characterized by: It includes a positioning and prediction unit (1), an analysis and exclusion unit (2), and an evaluation and delivery unit (3); The positioning and prediction unit (1) obtains the user's favorite advertisement data, the user's movement pattern and historical data and performs positioning, collects the user's facial expression image data and voice intonation data based on the positioning information, and then predicts the user's stay location based on the acquired historical data; The analysis and exclusion unit (2) is used to receive the user's favorite advertisements and the predicted user's stay location points obtained by the positioning and prediction unit (1), and perform a preference analysis of the favorite advertisements and identify low preference factors based on the obtained user's favorite advertisements, and then collect surrounding environment information based on the predicted user's stay location points, and then exclude low preference factor data from the collected surrounding environment information; The evaluation and delivery unit (3) is used to receive the user facial expression image data, voice and intonation data collected by the positioning and prediction unit (1) and the location point surrounding environment information collected by the analysis and exclusion unit (2), and to evaluate the user's emotions based on the collected user facial expression image data and voice and intonation data, and then to deliver the advertisement based on the evaluated user emotions and the collected location point surrounding environment information.

2. The advertising placement management system based on preference analysis according to claim 1, characterized in that: The positioning and prediction unit (1) comprises a positioning acquisition module (11) and a position prediction module (12); The positioning acquisition module (11) is used to acquire user-favorite advertising data, user movement patterns and historical data, acquire user location data in real time based on the acquired user movement patterns for positioning, and use surrounding cameras to acquire user facial expression image data and voice intonation data based on the positioning information; The location prediction module (12) is used to receive the positioning information, the acquired user movement pattern and the acquired historical data in the positioning acquisition module (11), acquire movement trajectory data from the user movement pattern data, and analyze the user's activity range based on the movement trajectory, and then use a random model to predict the user's stay location point based on the analyzed user activity range and the acquired historical data, and locate the user based on the predicted passing location point and use cameras around the predicted passing location point to capture the user's facial expression image data.

3. The advertising placement management system based on preference analysis according to claim 2 is characterized by: The analysis and exclusion unit (2) comprises an analysis and comparison module (21) and a data exclusion module (22); The analysis and comparison module (21) is used to receive the user's favorite advertisements obtained in the positioning collection module (11), and use association rule mining to perform a preference analysis of the favorite advertisements based on the obtained user's favorite advertisements, and then use the analyzed preference of the favorite advertisements to compare with the set preference of the advertisements; The data exclusion module (22) is used to receive the user's stay location predicted by the location prediction module (12) and the low preference factor data in the analysis and comparison module (21), perform positioning according to the predicted user's stay location, collect surrounding environment information based on the positioning information and using a camera, exclude the low preference factor data from the collected surrounding environment information, and obtain the excluded environment information.

4. The advertising placement management system based on preference analysis according to claim 3 is characterized by: The specific comparison conditions in the analysis and comparison module (21) include: Case ①: When the degree of preference of the analyzed favorite advertisement is greater than the set degree of preference of the advertisement, it indicates that the obtained user favorite advertisement has a high degree of preference, and a preference model is established based on the obtained user favorite advertisement. The preference model is used to collect the obtained user favorite advertisements, and the degree of preference of the obtained user favorite advertisements is higher than the set degree of preference of the advertisement. The preference model is used to classify the collected user favorite advertisements into advertising types. The types of favorite advertisements in the preference model are divided into humorous, healing, warm and cheerful. Case ②: When the preference of the analyzed preferred advertisement is less than the set advertisement preference, it indicates that the obtained user preferred advertisement has a low preference degree, and the low preference factor is identified based on the obtained user preferred advertisement.

5. The advertising placement management system based on preference analysis according to claim 3 is characterized by: The evaluation and delivery unit (3) is used to receive the user facial expression image data, voice and intonation data collected by the positioning and acquisition module (11), the user facial expression image data captured by the position prediction module (12), and the predicted user stay location point, receive the preference model in the analysis and comparison module (21) and the environmental information excluded by the data exclusion module (22), and the collected location point surrounding environmental information, and use a support vector machine to evaluate the user's emotions based on the collected user facial expression image data, voice and intonation data and the captured user facial expression image data, the evaluated user emotions are "pleasure" and "anger", and deliver advertisements based on the evaluated user emotions and the collected location point surrounding environmental information.

6. The advertising placement management system based on preference analysis according to claim 5, characterized in that: The specific delivery steps in the evaluation delivery unit (3) include: S1. When the user's emotion is evaluated as positive and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertisement types in the preference model, and the preferred advertisement types related to the excluded environmental information are matched, and then the preferred advertisement types are sorted according to the preference degree of the related preferred advertisement types; S2. When the evaluated user emotion is anger, and the user is about to reach the predicted user stay location, the excluded environmental information is used to match the advertising types in the preference model to match the preferred advertising types related to the excluded environmental information, and then a recurrent neural network is used to analyze the cause of anger based on the user's web browsing records within the analyzed activity range time, and the related preferred advertising types containing the analyzed cause of anger are excluded based on the analyzed cause of anger, and the related preferred advertising types excluding the cause of anger are obtained, and then humorous type advertisements are extracted from the related preferred advertising types excluding the cause of anger, and then the extracted humorous type advertisements are compared and ranked using the preference degree of the extracted humorous type advertisements.

7. The advertising placement management system based on preference analysis according to claim 6, characterized in that: The specific ranking of S1 in the evaluation and placement unit (3) includes: Case 1: If the ranked ads are ranked from low to high in terms of their likeability, the most likeable ad is selected and placed; Case 2: If the ranked advertisement preferences are from low to high, and there are two or more related preference advertisements with the highest advertisement preference, then the related preference advertisement with the highest advertisement preference is randomly selected for advertisement delivery.

8. The advertising placement management system based on preference analysis according to claim 6, characterized in that: The specific ranking of S2 in the evaluation and delivery unit (3) includes: Case 1: If the ranked advertisements are ranked from low to high in terms of their likeability, the humorous advertisement with the highest likeability is selected and placed; Case 2: If the ranked advertisement preference is from low to high, and there are two or more humorous type advertisements with the highest preference, the extracted humorous type advertisements are evaluated for humorousness using random forest, and the humorous type advertisement with the highest humorousness is selected for advertisement delivery.

9. The advertising placement management system based on preference analysis according to claim 6, characterized in that: The evaluation and delivery unit (3) S2 uses random forest to evaluate the humor level of humorous advertisements to achieve the following steps: Step 1: First, collect the extracted humorous advertisements and use them as training humorous data sets. Each humorous advertisement text is used as a humorous sample, and its humorous degree score is used as the corresponding label. Then, feature extraction is performed on the humorous samples. The word embedding model can be used to convert the text into a vector representation, and the feature word frequency and text length are extracted. Step 2: Use the training humor dataset to train the random forest regression model. The random forest constructs multiple decision trees and obtains the number of decision trees in the random forest. And the prediction results of the decision tree , It is The prediction results of a decision tree; Step 3: Humor training set ( ,in is a characteristic of humorous advertising texts, is the corresponding humor score, and then the prediction results of the decision tree are summed up to obtain the summed decision tree prediction results ; Step 4: Use weighted decision tree to predict results , the number of decision trees in the random forest And the prediction results of the decision tree Conduct a humor level assessment and obtain the humor level of the assessment , the specific algorithm formula is: 。

Images