Food advertisement intelligent analysis and promotion method and system based on big data

Abstract

The invention discloses a food advertisement intelligent analysis and promotion method and system based on big data. The prediction method comprises the following steps: collecting user data, collecting cross-platform food hotspots and emotional tendencies, collecting various food advertisements, and forming an advertisement library; the preprocessed data stream is input into a feature extraction module, a three-dimensional situation vector is extracted according to a situation vector engine, and a social emotion vector is extracted according to an emotion fluctuation measurement engine; inputting the three-dimensional situation vector and the social emotion vector into a demand mapper in parallel, generating a demand particle rule base, and detecting an optimization demand through a demand entanglement effect; the demand particle rule base retrieves a matched candidate promotion advertisement set from an advertisement base through a gravitation matching algorithm and emotion resonance intensity; and performing advertisement delivery through a delivery decision tree according to a matching result, and feeding back a delivery log to the evaluation module to update matching parameters. Advertisement promotion is carried out according to the matching degree of the advertisement and the user demand, and the accuracy of advertisement promotion is improved.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and in particular to a method and system for intelligent analysis and promotion of food advertising based on big data. Background Technology

[0002] Traditional advertising relies on demographic characteristics for targeting, making it difficult to capture real-time user needs. Furthermore, user behavior data is scattered across multiple channels, including e-commerce platforms, social media, and offline consumption, making integration and analysis challenging. This leads to outdated advertising strategies, wasted budgets on large-scale campaigns, and difficulty in quantifying ROI. User needs are becoming increasingly personalized, with niche demands such as healthy eating, regional specialties, and emotional consumption emerging. Traditional standardized advertising struggles to meet these diverse preferences. Therefore, this invention proposes a big data-based intelligent analysis and promotion method and system for food advertising. Summary of the Invention

[0003] This invention provides a method for intelligent analysis and promotion of food advertising based on big data, characterized by comprising:

[0004] S10: Collect user behavior, physiological state, and ambient heat map; collect cross-platform food hotspots and sentiment trends; preprocess multi-source real-time data streams; collect various food advertisements; and construct an advertisement library.

[0005] S20. The preprocessed data stream is input into the feature extraction module, which extracts a three-dimensional context vector based on the context vector engine and a social emotion vector based on the emotion fluctuation measurement engine.

[0006] S30. Input the three-dimensional context vector and the social emotion vector in parallel into the demand mapper to generate a demand particle rule base, and detect and optimize the demand through the demand entanglement effect.

[0007] S40. The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using the gravity matching algorithm and the intensity of emotional resonance.

[0008] S50. Based on the matching results, ads are delivered through the delivery decision tree, and the matching parameters are updated by feeding back the context fit and emotional resonance in the delivery log to the evaluation module.

[0009] The above-described intelligent analysis and promotion method for food advertising based on big data includes a context vector engine that extracts three-dimensional context vectors, which consists of the following sub-steps:

[0010] The geospatial area is quantified into a honeycomb grid, the commercial energy value of each grid is calculated, and the grid is matched according to the user's location;

[0011] Define a scenario-based time period rule base to match corresponding consumption time period tags based on user demand time.

[0012] Calculate the actual human perception parameters based on the user's location and status;

[0013] The system stitches together the user's grid location, consumption time period, and actual human perception parameters to output a three-dimensional context vector.

[0014] The above-described intelligent analysis and promotion method for food advertising based on big data includes a sentiment fluctuation measurement engine that extracts social sentiment vectors, which is divided into the following sub-steps:

[0015] Based on the frequency of positive, negative, and controversial words in food-related topics from the preliminary sentiment data set, the sentiment energy value of the topics is calculated.

[0016] The resonance intensity of the food theme is calculated based on the emotional energy value of the topic. The emotional dimension is set according to the common emotional themes and resonance intensity of the food. The social emotional vector is obtained by splicing the emotional dimensions.

[0017] The above-described intelligent analysis and promotion method for food advertising based on big data involves inputting a three-dimensional context vector and a socio-emotion vector in parallel into a demand mapper to generate a demand particle rule base. This method comprises the following sub-steps:

[0018] By using association rule mining and causal analysis, high-frequency, strongly correlated patterns are extracted from historical context-emotion combinations to construct an initial rule base.

[0019] By using game theory to balance the conflicts among the three parties and establishing a rule evolution mechanism through real-time feedback data, the rules can be adaptively evolved.

[0020] The above-described intelligent analysis and promotion method for food advertising based on big data, which establishes a rule evolution mechanism through real-time feedback data to achieve adaptive evolution of rules, specifically consists of the following sub-steps:

[0021] The sliding window weighted algorithm is used for rule strength iteration;

[0022] Mark them as conflict pairs and handle them differently depending on the strength of the conflict;

[0023] Injecting disaster scenarios to simulate extreme environmental stress tests.

[0024] The above-described intelligent analysis and promotion method for food advertising based on big data, which detects and optimizes demand through the demand entanglement effect, specifically includes the following sub-steps:

[0025] The basic requirement is to match particles with environmental signals to verify the entanglement effect;

[0026] The basic requirement particles are matched with the upgrade requirements in the upgrade rule base, and new upgrade requirement particles are generated based on the upgrade requirements.

[0027] The above-described intelligent analysis and promotion method for food advertising based on big data involves a demand particle rule base that retrieves a set of matching candidate advertisements from an advertising database using a gravity matching algorithm and emotional resonance intensity. Specifically, it comprises the following sub-steps:

[0028] The initial gravity value is calculated using a gravity matching algorithm based on the demand vector and the advertising vector;

[0029] The initial attraction value score is adjusted based on the intensity of emotional resonance in demand and the characteristics of emotional labels in advertising.

[0030] A set of candidate promotional ads is generated using a diversity optimization ranking algorithm and dynamic threshold filtering.

[0031] This invention also provides a big data-based intelligent analysis and promotion system for food advertising, comprising:

[0032] Data collection module: collects user behavior, physiological state, and ambient heat map; collects cross-platform food hotspots and sentiment trends; preprocesses multi-source real-time data streams; and collects various food advertisements to form an advertisement library.

[0033] Vector Extraction Module: The preprocessed data stream is input into the feature extraction module, which extracts a three-dimensional context vector based on the context vector engine and a socio-emotion vector based on the emotion fluctuation measurement engine.

[0034] Demand rule generation and optimization module: Input the three-dimensional context vector and social emotion vector in parallel into the demand mapper to generate a demand particle rule library, and detect and optimize demands through the demand entanglement effect;

[0035] The candidate set generation module: The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using the gravity matching algorithm and the intensity of emotional resonance;

[0036] The ad delivery and feedback module delivers ads based on the matching results through a decision tree and feeds back the contextual fit and emotional resonance in the delivery logs to the evaluation module to update the matching parameters.

[0037] The beneficial effects achieved by this invention are as follows: This invention promotes advertising based on the matching degree between advertisements and user needs, thereby improving the accuracy of advertising promotion. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0039] Figure 1This is a flowchart of a big data-based intelligent analysis and promotion method for food advertising, provided in Embodiment 1 of this application.

[0040] Figure 2 This is a schematic diagram of a big data-based intelligent analysis and promotion system for food advertising, provided in Embodiment 2 of this application. Detailed Implementation

[0041] 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, not all, of the embodiments of the present invention. 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.

[0042] Example 1

[0043] like Figure 1 As shown, Embodiment 1 of this application provides a method for intelligent analysis and promotion of food advertising based on big data, including:

[0044] S10. Collect user status and surrounding environment information, collect cross-platform food hotspots and sentiment trends, preprocess multi-source real-time data streams, collect various food advertisements, and form an advertisement library.

[0045] S11. Collect user behavior, physiological state, and surrounding environment based on the location information provided by the user.

[0046] Based on user-authorized location information, and utilizing sensors such as accelerometers and gyroscopes built into smart terminals, combined with inertial navigation algorithms, the system continuously tracks the user's movement trajectory, steps, activity intensity, and other motion behavior data. Through Bluetooth connections with wearable devices such as smart bracelets and smartwatches, it reads biosensor data in real time, accurately analyzing the user's physiological state, including heart rate variability and blood oxygen saturation. Simultaneously, it leverages geographic information systems and location services to obtain the user's location's temperature, humidity, and ultraviolet radiation, and analyzes the user's frequently used apps and food preferences based on software usage records.

[0047] S12. Cross-platform scraping of food-related content to obtain a preliminary data set of food trends and sentiment.

[0048] We efficiently extract food-related content from publicly available social media discussions, news comments, trending topics on short video platforms, real-time search engine trends, and the latest reviews and sentiments on food review websites. This content includes hot topics such as food production safety incidents, new food trends, updates on popular foods, and product order trends.

[0049] Trending food items are identified by analyzing order volume and category aggregation trends on food delivery platforms. Natural language processing (NLP) is used to preprocess the captured food text, including cleaning, word segmentation, and part-of-speech tagging. Then, using dictionary-based and grammatical rule-based methods, real-time analysis is conducted to identify macro-level social sentiment focus, the popularity of food-related topics, and sudden food-related trending events, thus recognizing collective needs and emotions. Simultaneously, factors such as cultural background, context, emotional intensity, emoticons, and keyword popularity are considered to improve the accuracy of retaining sentiment data related to trending foods. A preliminary sentiment data set is formed by combining trending food items and related sentiment trend data.

[0050] S13. Preprocess the multi-source real-time data stream.

[0051] The preprocessing steps include removing duplicate data, null values, and data with abnormal formatting, and detecting sudden outliers using a sliding window. A rule engine filters invalid data, retaining only valid feature fields. The format of multi-source data is standardized, converting unstructured data into structured fields. Unit conversion and data normalization are performed to ensure dimensional consistency. Data from different sources is synchronized based on timestamps, and delayed data is handled using a sliding window mechanism. Association rules are used to fuse multi-dimensional data from a unified region, the same food item, or the same topic.

[0052] S14. Collect food advertisements for various categories, label each food advertisement, and form an advertisement library.

[0053] We collect food advertisements across various categories and, based on the product's inherent attributes and potential emotional value, assign rich, multi-dimensional tags to each advertisement: Based on functional attributes, we categorize them as hot / cold, sweet / salty, portable / requires cooking, filling / snack, energizing / soothing, etc.; based on contextual relevance, we categorize them as breakfast / lunch / dinner / late-night snack, office / outdoor / home / commuting, etc.; based on emotional value tags, we categorize them as nostalgic, healing, novelty, social, health, luxury, convenience, reward, etc.; based on sensory association tags, we categorize them as crispy, silky, juicy, cool, warm, richly aromatic, etc.; and based on social tags, we categorize them as trendy, classic, environmentally friendly, local specialty, imported, etc. Based on the analysis of the social emotional field, we dynamically adjust the weight of the tags or add new tags.

[0054] In addition, each advertisement must clearly define its basic data, contextual data, sentiment data, and keywords in the advertisement library. The basic data includes the advertisement code, product category, material type, and delivery channel. The contextual data includes temperature, weather, geographic location type, time period, and whether it is a holiday.

[0055] S20. The preprocessed data stream is input into the feature extraction module, which extracts a three-dimensional context vector based on the context vector engine and a social emotion vector based on the emotion fluctuation measurement engine.

[0056] The preprocessed data stream is input into the feature extraction module, which includes a context vector engine and a sentiment fluctuation measurement engine.

[0057] S21. The context vector engine extracts the 3D context vector, which is divided into the following sub-steps:

[0058] S211. Quantize the geospatial area into a honeycomb grid, calculate the commercial energy value of each grid, and match the grid according to the user's location.

[0059] Using the Global Geodetic Coordinate System, the city's administrative boundary data in GeoJSON format was read, and the latitude and longitude coordinates were converted to a web Mercator projection to eliminate spherical distance errors. Based on a 1km... 2 For the area requirement, select the H3 resolution level, use the city center point as the seed, divide the area into hexagonal grids with a side length of 1km, expand radially, iterate until the city boundary is covered, and delete redundant grids where the center point is more than 15km away from the boundary.

[0060] Based on the H3 index, a spatial ID intelligent code is generated for each grid. The commercial energy value within each grid is calculated, and the grid's advertising placement strategy is analyzed based on the energy value rating. The formula for calculating the commercial energy value is as follows: Wherein, α·F(N shop ) represents the grid commercial density term. For the business density function, w i For store category weight, n i Where λ represents the number of stores in the corresponding product category, and D represents the strength of the new store gain. new P represents the number of new stores opened within a month, and c represents the gain saturation threshold, which determines the critical point affected by the number of new stores. real P represents the current real-time pedestrian flow. max The maximum historical pedestrian flow for this grid is represented by η, where η is the attenuation coefficient, set to 0.4 for weekdays and 0.2 for weekends. t represents the current time period. peak T represents the historical peak pedestrian flow period for this grid. day For the time span of a day, γ·H(W) trans () represents the traffic potential energy term. Let m be the traffic potential energy function, and k be the number of traffic hub types. j This is the coefficient for the type of transportation hub. Transportation hubs are divided into subway stations, bus hubs, shared bicycle stations, and parking lots. For the corresponding type count, δ is the decay exponent based on diminishing marginal utility, ε is the parking lot weight coefficient, and C park To find the number of parking spaces within a grid, divide the number of parking spaces by 100 to standardize it, converting it into a value in units of 100 parking spaces.

[0061] The system matches grids based on the user's location and records the business energy value of the grid.

[0062] S212. Define a scenario time period rule base and match the corresponding consumption time period tag according to the user's required time.

[0063] Based on the time dimension and the different food needs of different groups at different times, a day is divided into 12 consumption periods: 5:30-7:00 AM is the morning wake-up period; 7:00-9:00 AM is the breakfast rush period; 9:00-11:30 AM is the morning continuation period; 11:30-1:30 PM is the lunch period; 1:30-2:30 PM is the post-meal buffer period; 2:30-4:00 PM is the afternoon tea social period; 4:00-3:30 PM is the after-school replenishment period; 3:30-7:30 PM is the dinner pre-heating period; 7:30-9:00 PM is the dinner period; 9:00-11:00 PM is the overtime comfort period; 11:00-1:00 AM is the late-night emergency period; and 1:00-5:30 AM is the tail end of the night economy. Consumption periods are matched according to user needs and recorded.

[0064] S213. Calculate the actual human perception parameters based on the user's location and status.

[0065] The human body's actual perceived parameters consist of perceived temperature and exercise intensity. The perceived temperature is calculated based on the user's location, including temperature, humidity, and wind speed, using a thermal index algorithm. This perceived temperature is also affected by the user's exercise intensity. Exercise intensity is determined by extracting data such as metabolic rate, oxygen uptake, heart rate, and exercise speed from the exercise and physiological data obtained in step S10. This data, along with the user's subjective feelings, is used to comprehensively assess the exercise intensity. Both of these parameters must consider the differences in sensitivity among individuals.

[0066] S214, Concatenate and output the three-dimensional context vector.

[0067] The user's grid location, grid business energy value, consumption time period, and human body's actual perception parameters are stitched together to output a three-dimensional context vector.

[0068] S22. The emotion fluctuation measurement engine extracts social emotion vectors, which is divided into the following sub-steps:

[0069] S221. Calculate the emotional energy value of a food topic by statistically analyzing the frequency of positive, negative, and controversial words in the preliminary emotional data set.

[0070] Sentiment quantification was performed on the data in the preliminary sentiment dataset. Sentiment words were matched to comments and content under each food topic. A predefined sentiment dictionary was used to count the frequency of positive, negative, and controversial sentiment words. The frequency of positive, negative, and controversial sentiment words was weighted and calculated to obtain the sentiment intensity, which was then normalized to the [-1, 1] interval. Sentiment energy values ​​were calculated based on topic participation and the energy decay curve of sentiment intensity. The sentiment energy value reflects the sentiment bias of the food topic; an energy value of -1 indicates completely negative sentiment, and an energy value of 1 indicates completely positive sentiment.

[0071] S222. Calculate the resonance intensity of the food theme based on the emotional energy value of the topic, set the emotional dimension based on the common emotional themes and resonance intensity of the food, and splice the emotional dimensions to obtain the social emotional vector.

[0072] A time decay coefficient is introduced to give higher weight to recent topics. Food-related sentiment keywords are extracted based on word frequency-inverse document frequency, and keyword weights are assigned according to keyword proportions. Clustering algorithms are used to absorb related words of keywords and form several sentiment themes and corresponding word clouds. Topics that emerge simultaneously on multiple platforms are identified, and the correlation coefficient of sentiment energy between platforms under the same food theme is calculated to obtain the resonance intensity of the theme. Emotional dimensions are set based on common sentiment themes and resonance intensity of food, such as health anxiety, weight loss, nostalgia, and cost-effectiveness sensitivity. Foods with the same sentiment keywords are grouped into the corresponding sentiment dimension, and the order of food is sorted according to keyword weight. The sentiment dimensions are then concatenated into a socio-sentiment vector.

[0073] S30. Input the three-dimensional context vector and the socio-emotion vector in parallel into the demand mapper to generate a demand particle rule base, and detect and optimize demands through the demand entanglement effect. Specifically, this is divided into the following sub-steps:

[0074] S31. Through association rule mining and causal analysis, high-frequency strong correlation patterns are extracted from historical context-emotion combinations to construct an initial rule base.

[0075] Historical data is structured and stored using three-dimensional context vectors and socio-sentiment vectors. Consumer behavior data—purchase category, average order value, and purchase channel—are added as factual indicators, forming a data cube. A priori connection algorithm is used to scan the data cube based on pre-defined empirical rules, uncovering demand-related rules and extracting strong rule chains. The scanning condition is meeting the set minimum support and minimum confidence levels. A counterfactual intervention algorithm is used to perform counterfactual testing on the uncovered demand rules, comparing the change in target product sales within 3 hours before and after the rule trigger, retaining rules with an increase rate >15%. A topological clustering algorithm is used to merge similar demand rules, outputting an initial set of demand rules for the rule base, which are then stored in the rule base.

[0076] S32. Utilize game theory to balance the conflicts among the three parties, and establish a rule evolution mechanism through real-time feedback data to achieve adaptive evolution of the rules.

[0077] Based on a three-party game model, this study balances the conflicts among three parties: consumer utility, merchant profit, and platform ecosystem. By simultaneously solving the Pareto optimal solution of the consumer utility function, merchant profit function, and platform ecosystem value function, and iteratively calculating the Nash equilibrium point under the constraints of the three parties, the study ultimately outputs a demand strategy that maximizes overall benefits.

[0078] By establishing a rule evolution mechanism through real-time feedback data, the adaptive evolution of rules can be achieved, which consists of the following sub-steps:

[0079] S321. A sliding window weighted algorithm is used for rule strength iteration. The search growth rate of related categories after the rule is triggered is calculated every 24 hours, and the weights are dynamically adjusted according to the new formula. If the growth rate of a certain rule is greater than 15% for 3 consecutive days, the weight is increased by 20%.

[0080] S322. Traverse the initial rule base. When the overlap rate of the triggering conditions of two rules is >60% but the output demand direction is opposite, mark them as conflict pairs. Multiply the absolute value of the confidence difference between the two rules by the historical simultaneous triggering rate to obtain the rule conflict intensity. If the conflict intensity is weak, use weighted average to generate new demand. If the conflict intensity is strong, use demand fusion algorithm to generate new rules.

[0081] S323. By injecting disaster scenarios to simulate extreme environmental stress tests, invalid rules that fail to meet the standards for three consecutive tests are eliminated. When a food safety incident occurs, the rule topology is reorganized, all rules containing affected categories are disabled, backup rules are activated, and the rule tree is gradually rebuilt within 72 hours.

[0082] The final output is a dynamically optimized rule base.

[0083] S33. Optimize requirements by detecting the demand entanglement effect.

[0084] Detect the demand entanglement effect on the rule base. When two or more related demands occur simultaneously within the same spatiotemporal grid, they interact to generate a new, upgraded demand.

[0085] Real-time monitoring of demand particle triggering status is performed, recording the type, intensity, and triggering time of demand particles triggered within each grid. Based on a pre-defined list of entangled demand pairs, potential entangled demand combinations are identified. For each potential entangled pair, it is checked whether the intensities of both demand particles within the same grid, at the same time, exceed a threshold. If these conditions are met, the entanglement intensity is calculated as the product of the smaller of the change rates of the demand and environmental signals and a time overlap coefficient, which is the overlap period length divided by the total monitoring period. When the entanglement intensity exceeds a pre-defined threshold, a new upgraded demand particle is generated, and its subsequent transformation effect is recorded to adjust the entanglement rule weights. The generation of a new upgraded demand particle involves the following sub-steps:

[0086] S331. Basic requirement: Particles and environmental signals are matched to verify the entanglement effect.

[0087] Real-time monitoring of generated basic demand particles and specific environmental signals. When both exceed a preset threshold simultaneously within the same spatiotemporal grid, matching is triggered. If, within a short period, the rate of change of both significantly increases relative to the historical baseline, the Pearson correlation coefficient between the two signals is calculated: if the correlation coefficient > 0.85, indicating a strong positive correlation, then entanglement is confirmed.

[0088] S332. Match the basic requirement particles with the upgrade requirements in the upgrade rule base, and generate new upgrade requirement particles based on the upgrade requirements.

[0089] Based on a pre-defined upgrade rule library, the system matches the upgrade requirements corresponding to the combination of basic requirement particles and environmental signals, creating a new requirement particle object. This object includes: the upgraded requirement type, particle strength, a merged list of constraints, particle validity period, and a traceability flag. The strength of the new requirement particle is calculated by multiplying the strength of the basic requirement particle by a gain coefficient. After the upgrade particle is generated, its strength will decay over time or with changes in environmental signals. If the environmental signal falls below a threshold, the upgrade process is immediately terminated.

[0090] S40. The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using a gravity matching algorithm and emotional resonance intensity. This is specifically divided into the following sub-steps:

[0091] S41. Calculate the initial gravity value using the gravity matching algorithm based on the demand vector and the advertising vector.

[0092] Based on the vector space, demand rules in the demand particle rule base are converted into demand vectors, and ad tags in the ad library are converted into ad vectors. The demand vectors and ad vectors are then dimensionally aligned. For each ad vector in the ad library, the attraction value between it and the demand vector is calculated.

[0093] The formula is G is the context amplification term. base Based on the basic gravitational value, The context sensitivity coefficient. The magnitude of the three-dimensional context vector. Demand vector With ad vectors The similarity function, The magnitude of the demand vector. | represents the modulus of the ad vector. The overlap enhancement term produces an exponential boost when the key feature matching degree is high. κ represents the enhancement steepness, overlap is the proportion of matched key features, θ is the activation threshold, and Φ is the overlap enhancement term. constraints This is a constraint indicator. An indicator value of 0 indicates the existence of a hard conflict constraint (gravity value is zero), while an indicator value of 1 indicates no conflict. The gravity value is normalized to the range [0,1] for easier comparison.

[0094] S42. Adjust the initial attraction score based on the intensity of emotional resonance in demand and the characteristics of emotional labels in advertising.

[0095] For each ad, a cosine similarity algorithm is used to calculate the similarity between its sentiment tag features and the intensity of emotional resonance of the demand. An initial gravity value is weighted and combined with the sentiment similarity to generate an adjusted match score. If the sentiment similarity is below a threshold, the initial gravity value is penalized by multiplying it by a decay factor. The adjusted match scores of all ads are sorted and normalized to ensure a consistent score distribution.

[0096] S43. Use a diversity optimization sorting algorithm and dynamic threshold filtering to generate a set of candidate promotional ads.

[0097] A dynamic threshold is set based on the adjusted gravity score, and the top K percent of all ad scores are selected as candidates. This threshold is automatically adjusted according to the current score distribution, resulting in an initial candidate set. Ads in the initial candidate set are clustered by category label to form multiple groups. The cluster feature vector is the ad content feature vector. Within each cluster, ads are sorted according to the adjusted matching score, and the top M ads with the highest scores are selected from each group to ensure representation of each category. Simultaneously, to balance popular and unpopular categories, the value of M is dynamically adjusted based on the average popularity of ads within the group. The selected ads from each group are sorted from highest to lowest according to their adjusted matching score. If the total number exceeds the preset candidate set size N, the top N are selected. In the final candidate set, if multiple ads have similar scores, the sorting is fine-tuned based on their time period and relevance to demand, and the final candidate ad set is output.

[0098] S50. Based on the matching results, ads are delivered through the delivery decision tree, and the matching parameters are updated by feeding back the context fit and emotional resonance in the delivery log to the evaluation module.

[0099] The decision tree prioritizes ad placement within the candidate ad set based on the commercial energy value of the user's grid. First, high-energy grids prioritize high-priced, quality-oriented ads. Second, content strategies are differentiated based on user age: trendy new products are prioritized for younger users, while healthy traditional foods are prioritized for older users. Finally, a quantum annealing algorithm combined with a three-dimensional contextual analysis is used to find the optimal combination of promotional channels that meets the requirements, pushing ads to the most relevant contextual touchpoints, including specific software, digital screens near specific geographical locations, and information feeds at specific times. After each campaign, user feedback on contextual relevance and emotional resonance is collected. Contextual relevance is the increase in related searches in the area after ad exposure, and emotional resonance is the percentage of positive mentions on social media out of total mentions. Based on these two feedbacks, the evaluation module uses reinforcement learning to dynamically update the decision tree branch weights, while user interaction data is fed back to the S40 algorithm to optimize the gravity matching parameters.

[0100] Example 2

[0101] like Figure 2 As shown, Embodiment 2 of this application provides a food advertising intelligent analysis and promotion system based on big data, including:

[0102] The data collection module collects user behavior, physiological states, and ambient environmental heatmaps; it also collects cross-platform food trending topics and sentiment trends; preprocesses multi-source real-time data streams; and collects various food advertisements to form an advertising library. Specifically, it is divided into the following sub-modules:

[0103] User data submodule: Collects user behavior, physiological state, and surrounding environment based on the user's disclosed location information.

[0104] Food Data Submodule: Cross-platform scraping of food-related content to obtain a preliminary data set of food hot topics and sentiment trends.

[0105] Preprocessing submodule: preprocesses multi-source real-time data streams.

[0106] Advertising submodule: Collects food advertisements for various categories, tags each food advertisement, and forms an advertising library.

[0107] Vector Extraction Module: This module preprocesses the data stream input for feature extraction. It extracts a 3D context vector based on the context vector engine and a socio-emotional vector based on the emotion fluctuation measurement engine. Specifically, it is divided into the following sub-modules:

[0108] Context Vector Submodule: This module uses the context vector engine to extract 3D context vectors and is divided into the following submodules:

[0109] Grid Submodule: Quantizes geospatial space into a honeycomb grid, calculates the commercial energy value of each grid, and matches grids according to the user's location.

[0110] Time Period Submodule: Defines a rule library for scene time periods, matching the corresponding consumption time period tags according to the user's required time.

[0111] Perception Submodule: Calculates the actual human perception parameters based on the user's location and status.

[0112] The splicing submodule splices and outputs a three-dimensional context vector.

[0113] The social emotion vector submodule: This module uses an emotion fluctuation measurement engine to extract social emotion vectors, and is specifically divided into the following submodules:

[0114] Topic Energy Submodule: Calculates the topic sentiment energy value by statistically analyzing the frequency of positive, negative, and controversial words in food-related topics within the preliminary sentiment data set.

[0115] The calculation and splicing submodule calculates the resonance intensity of the food theme based on the emotional energy value of the topic, sets the emotional dimension based on the common emotional themes and resonance intensity of the food, and splices the emotional dimensions to obtain the social emotional vector.

[0116] Demand Rule Generation and Optimization Module: This module inputs a parallel concatenation of a 3D context vector and a socio-emotional vector into a demand mapper to generate a demand particle rule base. It then uses the demand entanglement effect to detect and optimize demands. Specifically, it consists of the following sub-modules:

[0117] The rule base construction sub-module extracts high-frequency, strongly correlated patterns from historical context-emotion combinations through association rule mining and causal analysis to construct the initial rule base.

[0118] The rule evolution submodule utilizes game theory to balance the conflict among the three parties and establishes a rule evolution mechanism through real-time feedback data to achieve adaptive rule evolution. Specifically, it consists of the following submodules:

[0119] Iterative Submodule: Employs a sliding window weighted algorithm to iterate rule strength, calculating the search growth rate of relevant categories after a rule is triggered every 24 hours and dynamically adjusting the weights according to the new formula. If a rule's growth rate exceeds 15% for three consecutive days, its weight is increased by 20%.

[0120] Conflict Submodule: Traverse the initial rule base. When two rules have an overlap rate of triggering conditions > 60% but output demand directions are opposite, they are marked as conflict pairs. The absolute value of the confidence difference between the two rules is multiplied by the historical simultaneous trigger rate to obtain the rule conflict strength. If the conflict strength is weak, a weighted average is used to generate new demands. If the conflict strength is strong, a demand fusion algorithm is used to generate new rules.

[0121] Environmental stress testing submodule: simulates extreme environmental stress testing by injecting disaster scenarios, eliminates invalid rules that fail to meet the standards after 3 consecutive tests, performs rule topology reorganization when a food safety incident occurs, disables all rules containing affected categories, activates backup rules, and gradually rebuilds the rule tree within 72 hours.

[0122] Optimization Submodule: This module detects and optimizes requirements through the requirement entanglement effect, and is specifically divided into the following submodules:

[0123] Verification submodule: Basic requirement: matching particles with environmental signals to verify entanglement effect.

[0124] Upgrade Submodule: Matches basic requirement particles with upgrade requirements in the upgrade rule library, and generates new upgrade requirement particles based on the upgrade requirements.

[0125] The candidate set generation module: The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using a gravity matching algorithm and emotional resonance strength. Specifically, it is divided into the following sub-modules:

[0126] Gravity Value Submodule: Calculates the initial gravity value using a gravity matching algorithm based on the demand vector and the advertising vector.

[0127] Adjustment submodule: Adjust the initial gravity value score based on the intensity of emotional resonance of demand and the emotional label characteristics of advertisement.

[0128] Generate Set Submodule: Uses diversity optimization sorting algorithms and dynamic threshold filtering to generate a set of candidate promotional ads.

[0129] The ad delivery and feedback module delivers ads based on the matching results through a decision tree and feeds back the contextual fit and emotional resonance in the delivery logs to the evaluation module to update the matching parameters.

[0130] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent analysis and promotion of food advertising based on big data, characterized in that, include: S10: Collect user behavior, physiological state, and ambient heat map; collect cross-platform food hotspots and sentiment trends; preprocess multi-source real-time data streams; collect various food advertisements; and construct an advertisement library. S20. The preprocessed data stream is input into the feature extraction module, which extracts a three-dimensional context vector based on the context vector engine and a social emotion vector based on the emotion fluctuation measurement engine. S30. Input the three-dimensional context vector and the social emotion vector in parallel into the demand mapper to generate a demand particle rule base, and detect and optimize the demand through the demand entanglement effect. S40. The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using the gravity matching algorithm and the intensity of emotional resonance. S50. Based on the matching results, ads are delivered through the delivery decision tree, and the matching parameters are updated by feeding back the context fit and emotional resonance in the delivery log to the evaluation module.

2. The intelligent analysis and promotion method for food advertising based on big data as described in claim 1, characterized in that, The context vector engine extracts 3D context vectors, which involves the following sub-steps: The geospatial area is quantified into a honeycomb grid, the commercial energy value of each grid is calculated, and the grid is matched according to the user's location; Define a scenario-based time period rule base to match corresponding consumption time period tags based on user demand time. Calculate the actual human perception parameters based on the user's location and status; The system stitches together the user's grid location, consumption time period, and actual human perception parameters to output a three-dimensional context vector.

3. The intelligent analysis and promotion method for food advertising based on big data as described in claim 1, characterized in that, The emotion fluctuation measurement engine extracts social emotion vectors, which is divided into the following sub-steps: Based on the frequency of positive, negative, and controversial words in food-related topics from the preliminary sentiment data set, the sentiment energy value of the topics is calculated. The resonance intensity of the food theme is calculated based on the emotional energy value of the topic. The emotional dimension is set according to the common emotional themes and resonance intensity of the food. The social emotional vector is obtained by splicing the emotional dimensions.

4. The intelligent analysis and promotion method for food advertising based on big data as described in claim 1, characterized in that, The three-dimensional context vector and the socio-emotion vector are input in parallel into the demand mapper to generate a demand particle rule base, which is divided into the following sub-steps: By using association rule mining and causal analysis, high-frequency, strongly correlated patterns are extracted from historical context-emotion combinations to construct an initial rule base. By using game theory to balance the conflicts among the three parties and establishing a rule evolution mechanism through real-time feedback data, the rules can be adaptively evolved.

5. The intelligent analysis and promotion method for food advertising based on big data as described in claim 4, characterized in that, By establishing a rule evolution mechanism through real-time feedback data, the adaptive evolution of rules can be achieved. This can be divided into the following sub-steps: The sliding window weighted algorithm is used for rule strength iteration; Mark them as conflict pairs and handle them differently depending on the strength of the conflict; Injecting disaster scenarios to simulate extreme environmental stress tests.

6. The intelligent analysis and promotion method for food advertising based on big data as described in claim 1, characterized in that, Detecting and optimizing requirements through the demand entanglement effect involves the following sub-steps: The basic requirement is to match particles with environmental signals to verify the entanglement effect; The basic requirement particles are matched with the upgrade requirements in the upgrade rule base, and new upgrade requirement particles are generated based on the upgrade requirements.

7. The intelligent analysis and promotion method for food advertising based on big data as described in claim 1, characterized in that, The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using a gravity matching algorithm and emotional resonance intensity. Specifically, it consists of the following sub-steps: The initial gravity value is calculated using a gravity matching algorithm based on the demand vector and the advertising vector; The initial attraction value score is adjusted based on the intensity of emotional resonance in demand and the characteristics of emotional labels in advertising. A set of candidate promotional ads is generated using a diversity optimization ranking algorithm and dynamic threshold filtering.

8. A smart analysis and promotion system for food advertising based on big data, characterized in that, include: Data collection module: collects user behavior, physiological state, and ambient heat map; collects cross-platform food hotspots and sentiment trends; preprocesses multi-source real-time data streams; and collects various food advertisements to form an advertisement library. Vector Extraction Module: The preprocessed data stream is input into the feature extraction module, which extracts a three-dimensional context vector based on the context vector engine and a socio-emotion vector based on the emotion fluctuation measurement engine. Demand rule generation and optimization module: Input the three-dimensional context vector and social emotion vector in parallel into the demand mapper to generate a demand particle rule library, and detect and optimize demands through the demand entanglement effect; The candidate set generation module: The demand particle rule base retrieves a set of matching candidate promotional ads from the ad library using the gravity matching algorithm and the intensity of emotional resonance; The ad delivery and feedback module delivers ads based on the matching results through a decision tree and feeds back the contextual fit and emotional resonance in the delivery logs to the evaluation module to update the matching parameters.

Images