Multi-modal behavior emotion perception edge computing regulation and control system

Abstract

The invention relates to a multi-modal behavior emotion perception edge calculation regulation and control system, and belongs to the technical field of edge calculation and behavior science crossing. Aiming at the technical problem of experience deterioration caused by continuous pushing of an existing advertisement system in a user emotion valley period, the method comprises the following steps: fusing a gyroscope pressure index, voice emotion polarity and consumption behavior sequence data in real time through terminal equipment, and generating an emotion index at an edge computing node; triggering a three-level permission state machine (LOCKED / BASIC / PRMIUM) based on an emotion index threshold value, and dynamically controlling the on-off and priority of an advertisement pushing channel; and meanwhile, a technical binding mechanism of the emotion index and a financial accounting system is established, and the emotion coefficient is written into an ISO 20022 message extension field to realize income distribution linkage. Tests show that the system reduces the advertisement complaint rate by 63%, improves the re-purchase rate by 41% after care, and solves the problem of accurate regulation and control of the service flow of the psychological dimension.

Description

I. TECHNICAL FIELD

[0001] The present application belongs to the technical field of edge computing and behavioral science, and specifically relates to an edge computing regulation system based on multi-modal behavioral emotion perception, which is particularly suitable for precise dynamic management of local life services. II. BACKGROUND

[0002] 2.1 There are three defects in the prior art:

[0003] 1. Traditional advertising systems (such as Google Ads) rely only on historical behavior data and cannot perceive real-time emotional states;

[0004] 2. Emotional computing advertising systems (such as IBM Watson CN107609572B) only achieve emotion recognition and advertising content matching, lacking a permission switch mechanism;

[0005] 3. User care systems (such as ByteDance's warm-hearted advertising) separate care actions from advertising push, and do not form a commercial closed loop.

[0006] 4. The present application breaks through the spatial dimension limitation and creates a technical path of "behavioral emotion perception → edge service regulation → financial sub-accounting → commercial closed loop", realizing service flow regulation in the psychological dimension.

[0007] 2.2 CROSS-REFERENCE STATEMENT

[0008] This patent is implemented in coordination with the following associated patent technology solutions of the same applicant submitted on the same day, which together constitute a complete technical framework: "Dynamic Geofence Precision Advertising System Based on Attributes and Values" (application in progress)

[0009] "Three-factor dynamic shopping coupon generation method and system based on circulation attenuation regulation" (application in progress)

[0010] For the complete solution, please refer to:

[0011] This patent: Figure 10

[0012] Dynamic Geofence Precision Advertising System Based on Attributes and Values Patent: Figure 6

[0013] Three-factor dynamic shopping coupon generation method and system based on circulation attenuation regulation patent: Figure 5

[0014] 2.3 Patent Coordination Protection Mechanism:

[0015] This patent and the associated patents "Dynamic Geofence Precision Advertising System Based on Attribute and Value", "Three-factor Dynamic Coupon Generation Method and System Based on Circulation Attenuation Adjustment" achieve:

[0016] 1. Technical indivisibility: any missing field leads to business loop failure

[0017] 2. Anti-avoidance design: separate implementation of extended fields cannot achieve complete technical effect

[0018] 3. Joint protection: three patents jointly cover the "emotion perception → service regulation → financial distribution" technical chain 2.4 Invention content

[0019] 2.4.1 Core innovation Figure 1 ) System architecture

[0020] 1. Data dimension breakthrough: from location data to three-source fusion of gyroscope stress indicators, voice emotion polarity, and consumption behavior sequence;

[0021] 2. Control mechanism transition: from advertisement screening to three-level state machine control of full-category service circulation;

[0022] 3. Business loop construction: emotion index generates ISO 20022 distribution message through edge encryption, realizing emotion coefficient financial penetration.

[0023] 2.4.2 Technical effect

[0024] 1. Advertisement complaint rate decreased by 63% (compared to traditional geofencing systems);

[0025] 2. Repurchase rate increased by 41% within 7 days after care touch;

[0026] 3. Multi-modal misjudgment rate decreased by 42% (behavior weight formula contributed 27% error reduction). III. BRIEF DESCRIPTION OF DRAWINGS

[0027] [ Figure 1 ] System architecture diagram: shows emotion calculation engine → permission decision tree → financial distribution linkage → business loop process Figure 2 Advertisement control strategy logic based on user emotion index calculation

[0028] [ Figure 3 ] Emotion index advertisement-business value closed loop

[0029] [ Figure 4 ] Dynamic three-level coupon regulation mechanism of emotion index

[0030] [ Figure 5 ]Emotional Care Execution Flow

[0031] [ Figure 6 ]Emotion-Ad Dynamic Decision Making Implementation System, Driving Precision Decision Making and Forming a Closed Loop

[0032] [ Figure 7 ]Self-evolution Decision System

[0033] [ Figure 8 ]Double Threshold Bidding Engine Closed Loop Process

[0034] [ Figure 9 ]Geographical Range Regulation Engine (LBS Dynamic Engine)

[0035] [ Figure 10 ]Full System Technology Linkage Business Closed Loop Process Based on Three Patent Collaborations

[0036] [ Figure 11 ]Emotion Coefficient Sharing Penetration Flowchart

[0037] [ Figure 12 ]Privacy First Data Processing and Permission Decision Flowchart

[0038] [ Figure 13 ]Financial Message Field Synergy Diagram IV. DETAILED DESCRIPTION

[0039] 4.1 Input Source and Processing Flow

[0040] 1. Multi-modal data collection:

[0041] Physiological signals: mobile phone gyroscope shaking frequency (stress indicator, frequency > 5Hz is determined as high pressure state)

[0042] Voice features: voice captured by microphone through MFCC feature extraction (dimension = 40)

[0043] Behavior sequence: consumer behavior log, including click timestamp and behavior type weight, weight

[0044] 1. Multi-modal behavior emotion perception implementation:

[0045] The module generates an emotion index through triple data fusion:

[0046] (a) Physiological stress perception: when the gyroscope angular velocity variance σ^2 > 0.05 rad^2 / s^2, P = 1.0, otherwise P = \frac{\sigma^2}{0.05}

[0047] (b) Voice Emotion Analysis: The LSTM model outputs the emotion polarity V∈[-1,1], which is then normalized using the formula V norm = (V+1)×50, converted to 0-100 points. (c) Calculation of consumption intention: using the time-series decay formula Wk=1 / (1+e - 0.1tk) weighted behavior sequence

[0048] Intent strength: I = Imax∑k=1n(Wk·wk)×100 where tk is the number of minutes after the action occurs. Constraints: (Maximum 5 valid actions in the last 30 minutes) wk∈[0.1,1.0](preset by merchant type (catering = 0.8, retail = 1.0);

[0049] Final output: Sentiment Index Formula

[0050] Constraints:

[0051] Normalization formula: V norm =V max -V min VV min ×100

[0052] Confidence decay function: μ=1 / (1+e - 5(δ-0.2))

[0053] The formula for the data difference rate is: δ=(max(P,Vnorm,I)-min(P,Vnorm,I)) / max(P,Vnorm,I)

[0054] The coefficients \alpha, \beta, \gamma are stored in the ARM TrustZone secure area using SM4 encryption.

[0055] 4.1.2 Definition of Technical Foundation: The technical foundation of this system is defined as a three-tier core architecture.

[0056] 1. Multimodal data fusion:

[0057] Definition: Spatiotemporal alignment and feature-level fusion of P, V, and I source data are completed at the device end.

[0058] Innovation: Introducing a time-series weight Wk with a decay factor of lambda = 0.1 (Wk = 0.953 at t = 30 min).

[0059] 2. Behavioral and emotional perception:

[0060] This refers to the analysis of intent intensity based on the time-series weighting of consumer behavior, with a decay factor λ = 0.1 controlling real-time sensitivity.

[0061] Ensure recent behavior gets higher weight (W=1.0 at t=0, W=0.95 at t=30min)

[0062] Technical features: I value calculation uses decay rate Ensure recent behavior dominates intent analysis (decay rate <5% at t<10min)

[0063] 3. The consumption intent weight formula is decayed by time factor e -0.1t Ensure recent behavior gets higher weight (W=1.0 at t=0, W≤0.05 at t≥30min), and superimpose merchant type preset weight to achieve precise intent capture.

[0064] 4. Edge computing regulation:

[0065] Refers to the completion of (i) emotional index calculation (ii) three-level permission decision (iii) account coefficient mark generation in the terminal device ARM TrustZone security area or 5G MEC node, the survival period of raw sensor data ≤50ms and never leave the physical boundary of the device.

[0066] 5. Patent synergy mechanism:

[0067] This system as the core engine of emotional computing, in-depth synergy with the following associated patents:

[0068] "Dynamic geographic fence precision advertising system based on attributes and value": receive the emotional index output by this system, dynamically adjust the geographic fence range (activate member expansion distance when emotion > 50)

[0069] "Three-factor dynamic shopping coupon generation method and system based on circulation decay adjustment": write the emotional coefficient as the shopping coupon inflation factor into the ISO 20022 message.

[0070] Synergistic data flow interacts in real time through gRPC protocol, timeout threshold 500ms, see Figure 10 .

[0071] Technical explanation:

[0072] In intent intensity calculation:

[0073] Merchant type weight wkwk is preset by business rules (dining = 0.8, retail = 1.0)

[0074] Time decay factor λ=0.1 λ=0.1 ensures that behavior weight decays to less than 5% within 30 minutes (W(t=30)≈0.05 W(t=30)≈0.05) Final intent score II is mapped to 0-100 points by weighted sum and normalization

[0075] 4.1.3 Algorithm implementation (python)

[0076]

[0077]

[0078] 4.1.4 Implementation of Consumer Behavior Weight Calculation (Patented Core Algorithm)

[0079]

[0080] 4.1.5 Post-consumption emotional correction system:

[0081]

[0082] 4.1.6 Confidence calibration mechanism:

[0083]

[0084] Technical results: The false positive rate decreased by 42% (verified by test cases), with the error rate contributed by the behavioral weight formula decreasing by 27% (compared to the static weight scheme).

[0085] 4.2 Affective Computing Engine

[0086] 4.2.1 Table 1 below shows the advertising placement control strategies based on user sentiment index calculation:

[0087]

[0088] Table 1

[0089] 4.2.2 Ad Placement Control Decision Logic Based on User Sentiment Index Figure 2 ).

[0090] Dynamic quota algorithm:

[0091] Python

[0092] def ad_quota_calc(emotion_index,member_level):

[0093] base_quota = [0, 0, 0] # Base quota for each level of emotion

[0094] level_bonus = member_level * 5 # Add 5 bonuses per level

[0095] return min(base_quota[emotion_tier]+level_bonus,35)#The upper limit is 35 items

[0096] Implementation code of permission decision tree

[0097]

[0098] 4.3 Emotion Index Ad-Commercial Value Closed-Loop Decision Tree Figure 3 )

[0099] 4.3.1: LTV Dynamic Pricing Formula

[0100]

[0101]

[0102] Emotion Threshold Elasticity Mechanism, Care Event Automatic Expansion Interval:

[0103]

[0104] Member Coefficient Calculation:

[0105]

[0106] 4.3.2 Technical Explanation:

[0107] When the Emotion Index is in the 30-70 interval, the system dynamically adjusts the number of basic ad slots according to the following formula: AdCount = BaseSlot x (EmotionIndex - 30) / 40

[0108] Key Algorithm, Emotion-Ad Matching Formula:

[0109] MatchScore = a · EmotionSimilarity + b · ConsumptionIntentMatchScore Where:

[0110] a = 0.7 (Emotion Similarity Weight)

[0111] b = 0.3 (Consumption Intent Weight)

[0112] Innovation: When \text{EmotionSimilarity} < 0.5, forcibly terminate ad push 4.3.3 Dynamic Weight Adjustment Rule Formula:

[0113]

[0114] 4.4 Dynamic Three-Level Coupon Competition Regulation Mechanism Based on Emotion Index, the user's emotional state is divided into three levels, and different economic strategies are used for each level. 1. The following table is a linkage example: set different trigger conditions (active click, display duration, care message) linkage according to different emotional states.

[0115]

[0116] Table 2

[0117] 4.4.1 Emotional resilience redemption formula

[0118] Actual freeze = benchmark freeze x (1-0.2x max(0, emotional index-70) / 30)

[0119] Honor reissue decay model

[0120]

[0121] 4.4.2 Three-level dynamic regulation diagram of coupon grabbing emotional index Figure 4 )

[0122] Emotional state perception algorithm:

[0123]

[0124]

[0125] Core algorithm implementation:

[0126]

[0127]

[0128] 4.4.3 Emotional care execution flowchart Figure 5 )

[0129] "Emotional data and bank-managed contribution value sharing system are physically isolated, and the advertising system only outputs permission instructions to the financial middle office" trigger conditions (user preset events):

[0130] Emotional index <40 for 3 consecutive days (system forced trigger)

[0131] Birthday / anniversary (need to be set by the user) 24 hours in advance to detect emotions.

[0132] 4.5 Emotional-advertising dynamic decision-making implementation system, driving accurate decision-making and forming a closed loop Figure 6 )

[0133] 4.5.1 Technical features

[0134] Ad conversion contribution value formula:

[0135] Contribution value = ad base value x emotional coefficient x actual conversion rate / industry benchmark

[0136] LBS fusion innovation engine

[0137]

[0138] Permission decision tree implementation logic

[0139]

[0140]

[0141]

[0142] 4.5.2 Multi-modal data fusion enhancement

[0143] Dynamic decay factor: Introduce time decay function to improve real-time performance

[0144]

[0145] 4.5.3 Self-evolving decision system Figure 7 ):

[0146] Self-evolving decision system automatically adjusts weights every week: New weight = old weight * (1 + strategy effect gain rate) Anti-cheating protection mechanism

[0147] Geofence drift detection: if (moving speed > 120 km / h) then freeze mood update

[0148] Emotion mutation monitoring: When the mood fluctuation is > 50 points within 1 hour, start manual review

[0149] Contribution value anomaly: Trigger risk control when single-day contribution value growth > threshold

[0150]

[0151]

[0152] 4.6 Double threshold bidding engine closed-loop flowchart Figure 8 )

[0153] 4.6.1 Double threshold bidding mechanism

[0154]

[0155] Double threshold bidding engine (patent core code)

[0156]

[0157]

[0158] Dynamic pricing premium stacking formula:

[0159] Final price = base price × mood premium coefficient × intention stacking coefficient

[0160] Mood index > 80 → premium coefficient 1.5

[0161] Consumption intention > 90 → overlay coefficient 1.2

[0162] Security communication protocol: OAuth2.0 JWT token verification

[0163] 4.6.2 Geographical range regulation engine (LBS dynamic engine)

[0164] This engine realizes the core demand of "advertising geographical range restriction" in the business model, and dynamically adjusts the push range through three factors:

[0165]

[0166]

[0167] Threshold decision rule (right item 5 / 13 protection point):

[0168] Emotion index ≤ 50: Member expansion value α × Lv forced to zero

[0169] Emotion index < 50: Geofence range × 0.7 contraction

[0170] Emotion index > 50: Activate member expansion function

[0171] Algorithm implementation:

[0172]

[0173] 4.7 Example

[0174] 4.7.1: Cross-patent technology collaboration generates core value flow:

[0175] 1. Emotion index → Geofence expansion → Improve advertising coverage

[0176] Emotion > 50 activates member expansion function (VIP 8 level maximum expansion 4KM)

[0177] Coverage improvement verification: Test scenario 3 shows that the reach radius is from 2.1KM → 4.5KM (+114%) 2. Emotion factor → shopping coupon inflation → improve conversion rate

[0178] Emotion > 70 emotion_factor = 1.5 triggers 150% inflation

[0179] Conversion rate improvement verification: Test case 2 shows that CTR is from 1.2% → 7.8% (+550%)

[0180]

[0181]

[0182] Table 3

[0183]

[0184] Scenario 1:

[0185] Advertisement regulation when user Ms. Wang breaks down

[0186] a. Data input:

[0187] Gyroscope data: shaking frequency 6.2 Hz (stress indicator ↑)

[0188] Voice analysis: "Don't want to see any promotions" (52% negative words)

[0189] b. System response:

[0190] Immediately turn off all advertisement pushes

[0191] Trigger care protocol: "Do you need a break? XX Coffee Shop offers free meditation space" (non-advertisement)

[0192] c. Business effect:

[0193] Advertisement complaint rate decreased by 63%

[0194] 7-day repurchase rate after care outreach increased by 41%

[0195] Note: The care protocol triggered in this scenario belongs to non-commercial push (permission status LOCKED) and does not trigger the advertisement bidding process.

[0196] Scenario 2:

[0197] VIP (Level 5) views dining advertisements when emotional index is 60

[0198] Basic range: 3KM

[0199] Member extension: (5-2) x 500m = 1.5KM

[0200] Actual range: 3 + 1.5 = 4.5KM

[0201] Scenario: Same user at emotional index 40:

[0202] Actual range: 3 x 0.7 = 2.1KM (Member extension ignored)

[0203] Scenario 3:

[0204] Scenario: VIP (Level 5) at shopping center (dining industry) emotional index 65:

[0205] Basic range = 3KM

[0206] Rank Extension = (5-2) x 500m = 1.5KM

[0207] Actual Range = 3 + 1.5 = 4.5KM

[0208] Push Result: Precise display of restaurant ads within 4.5KM

[0209] Scenario: Same user's emotional index 45:

[0210] Actual Range = 3 x 0.7 = 2.1KM

[0211] Push Result: Selected restaurant ads within 2.1KM are visible

[0212] Scenario 4:

[0213] System-wide coordination verification in high emotional state of VIP users: see high-value user service flowchart (10) based on multi-patent coordination.

[0214]

[0215]

[0216] Dynamic geographic fence precision advertising system based on attributes and value: Dynamic radius = 8KM + (8-2) x 500m = 11KM

[0217] (This patent):

[0218] Dynamic pricing calculation steps:

[0219] ① LTV factor: min(5000 / 1000, 20) = 5 → 5 x 0.5 = 2.5

[0220] ② Emotional time factor: (85 / 100) x 1.5 = 1.275 → 1.275 x 0.3 = 0.3825 / / Emotional 85 normalization

[0221] ③ Time period premium: max(1.5-1, 0) = 0.5 → 0.5 x 0.2 = 0.1

[0222] ④ Base price = 2.5 + 0.3825 + 0.1 = 2.9825 yuan

[0223] ⑤ Member coefficient superposition (VIP 8 levels): 2.9825 x (1 + 0.05 x 8) = 4.1755 yuan

[0224] - The proportion of coupon freezing is reduced to 80% (main right item g)

[0225] "Three-factor dynamic shopping voucher generation method and system based on flow attenuation adjustment patent: fund sub-account marker emotion_factor = 1.5"

[0226]

[0227] Technical description: #LBS dynamic engine algorithm (completely consistent with right item 13)

[0228]

[0229]

[0230] 4.7.2 Full system technology linkage business closed-loop process based on three patents (collaboration) Figure 10 )

[0231] Note: emotion_factor = 1.5 in the figure corresponds to > 70 emotional state defined by emotional patent

[0232] This graph shows the full system technology linkage in the high-value user scenario, and the specific process is as follows:

[0233] 1. User state input: position coordinates, emotion index (85), member level (8) as joint input, triggering attribute and value-based dynamic geographic fence precision advertising system patent calculation.

[0234] 2. Attribute and value-based dynamic geographic fence precision advertising system patent: calculate dynamic fence (right 1 protection point)

[0235] Execute the formula: R_dyn = min(industry benchmark radius + expansion factor x member level, maximum radius)

[0236] Output: 11KM precision push range

[0237] 3. Advertising system: call this patent for dynamic pricing request, (LTV = 5000 yuan)

[0238] 4. This patent: realize dynamic pricing (7.1 section) min(20,5) x 0.5 + (0.9 x 1.5) x 0.3 + max(1.5-1,0) x 0.2 = ￥8.6 yuan,

[0239] Coupon freezing decision (right g): emotion > 70 → freeze ratio down to 80%

[0240] 5. Three-factor dynamic shopping voucher generation method and system based on flow attenuation adjustment patent: write emotion coefficient into bank sub-account instruction, complete financial penetration (right 6)

[0241] Receive emotion_factor flag (emotion_factor = 1.5)

[0242] Generate ISO 20022 standard split account instruction (with extended field)

[0243] 6. Bank system: Receive instruction with emotion_factor through ISO 20022 message, distribute income to corresponding sub-account according to emotion_factor.

[0244] Parse emotion_factor = 1.5 flag

[0245] Distribute 1500 yuan of income to sub-account A (1.5 times the income pool)

[0246] Cross-patent communication protocol:

[0247] 1. Call specification: Through gRPC bidirectional streaming, the timeout threshold is set to 500 ms

[0248] 2. Data carrier:

[0249] Geofencing request: `{industry, member_level, emotion_index}`

[0250] Shopping voucher generation request: `{user_id, LTV_score, emotion_factor}`

[0251] 3. State synchronization: Exchange heartbeat packets every 5 minutes (including ε-difference noise parameters of each system)

[0252] 4.7.3 Financial penetration verification scenario

[0253]

[0254]

[0255] 4.8 Technical monetization model

[0256] 4.8.1 Paid advertising pricing formula

[0257] Price = Value Anchor (min(, 20) x 0.5) + Emotion Time Alignment (EmoScore x TimeFactor) x 0.3 + Gold Period Premium (max(TimeFactor-1, 0) x 0.2

[0258] Technical monetization model diagram

[0259]

[0260] Table 3

[0261] 4.8.2 Emotion Factor Rebate Penetration Flowchart Figure 11 )

[0262] The system realizes the financial-level binding of advertising revenue and emotional state:

[0263] Dynamic rebate: switch revenue multiples according to real-time emotion index (>70: 1.5 times, 30-70: 1.0 times)

[0264] Instruction generation: send rebate request to bank system through ISO 20022 financial message standard:

[0265]

[0266]

[0267] Factor penetration: write emotion factor label (emotion_factor) in the ExtensionField field of the message

[0268] Fund isolation: different multiple revenues enter independent sub-accounts to ensure that the merchant's 60% distribution rule can be audited ISO20022 message extension field implementation (right item 6 protection point):

[0269]

[0270] 4.8.3 Privacy Protection Data Processing and Permission Decision Flowchart Figure 12 )

[0271] Declare epsilon value: epsilon = 0.3 (meet strict GDPR standard)

[0272] Emotion index adds ±5 random noise (meets epsilon-differential privacy)

[0273] Ad permission state is encrypted as HMAC-SHA256 digest

[0274] Regulatory adaptation: original sensor data never leaves the device security zone

[0275] Comply with GDPR Article 25 "default data protection design"

[0276] 4.8.3.1 The system implements a privacy-first data processing paradigm

[0277] a. Device security zone isolation: original sensor data is physically isolated through ARM TrustZone technology

[0278] b. Edge computing node: encrypted data completes emotion index calculation in local / 5G MEC node

[0279] c. Permission decision mechanism: the output of the LOCKED / PREMIUM instruction triggers the corresponding business action

[0280] d. Zero raw data transmission: the entire process complies with the "default privacy design" principle of GDPR Article 25

[0281] 4.8.3.2 Technical Description:

[0282] The system realizes physical isolation of raw data through the device security zone, and only transmits to the edge node after SM4 encryption.

[0283] The system realizes end-to-end privacy protection:

[0284] 1. Raw data is processed and destroyed in the device security zone (left)

[0285] 2. Encrypted data is added with noise in the edge node (middle)

[0286] 3. Desensitized data generates business value in the cloud (right)

[0287] 4.8.3.3 Data Privacy Compliance Design:

[0288] 1. All emotional data is calculated on the device side (edge computing)

[0289] 2. Only upload the emotional index result to the server (original sensor data is not stored)

[0290] 3. Data minimization principle

[0291] Only collect the data necessary to realize the permission control of advertising (emotional index), do not collect raw biological features

[0292] 4. Default privacy design

[0293] ε=0.3 differential privacy meets the requirements of GDPR Article 25 (EDPB Guide WP265)

[0294] 5. Cross-border transmission guarantee

[0295] Through SM4 encryption + edge computing, realize data localization processing, avoid cloud storage cross-border risk

[0296] 6. User rights protection

[0297] Physical emergency button provides user immediate control (GDPR Article 22)

[0298] 4.8.3.4 The system ensures user data security through four levels of protection, realizing the core principle of "raw sensor data never leaves the device": 1. Device security zone isolation mechanism

[0299] Hardware-level Secure Enclave stores raw sensor data

[0300] Access control: only authorized for sentiment computing engine to read, prohibit network transmission module access

[0301] Technical standards: in line with GlobalPlatform TEE v2.0 specification

[0302] 2. SM4 national secret encryption process

[0303]

[0304]

[0305] 3. Edge computing node processing

[0306] Deployment location: device local / 5G MEC edge node

[0307] Processing content:

[0308] Sentiment index calculation (0-100)

[0309] Permission decision generation (LOCKED / BASIC / PREMIUM)

[0310] Output limit: only transmit decision instructions and desensitization index

[0311] 4. Differential noise addition (epsilon-differential privacy)

[0312]

[0313] 4.8.5 Financial message field coordination constraints Figure 11 )

[0314] The generation of the extended field of the ISO 20022 message needs to meet the cross-patent technology dependence:

[0315] `emotion_factor` (output of the system right item 6):

[0316] Write condition: advertisement permission state ≠ LOCKED (main right item S3.2)

[0317] Value range: 1.0 (emotional index 30-70) or 1.5 (> 70)

[0318] `geo_radius` (output of the main right item 1 of the geographic fence system):

[0319] Effective condition: emotional index > 50 (right item 13)

[0320] Outlier: ≤ 50 is forced to R_base x 0.7

[0321] `coupon_inflation` (shopping coupon system sovereign item 1 output):

[0322] Input validation: emotional index must be valid (shopping coupon system right item 8)

[0323] Degradation rule: default 1.0 when emotional interface is abnormal

[0324] 4.9 Emotional calculation engine

[0325] 4.9.1 Input data:

[0326] Gyroscope jitter frequency (stress indicator): sampling rate ≥ 50 Hz, frequency > 5 Hz is determined as high pressure Voice tone analysis (emotional polarity): use open source library Librosa to extract MFCC features, input LSTM model

[0327] Consumption behavior sequence (intention strength): 30-minute click stream weighted calculation (weight formula: W = 1 / (1+e^(-0.1*t)), t is time decay)

[0328] Technical feature comparison table

[0329]

[0330] Table 4

[0331] Algorithm:

[0332]

[0333] 4.9.2 Permission decision tree

[0334] State machine logic:

[0335]

[0336]

[0337] 4.9.3 Care event linkage, data flow verification

[0338] Database record example (avoiding privacy risks)

[0339]

[0340] 4.9.4 LBS dynamic engine technology description

[0341] Input interface:

[0342] 1. Emotion Index Pipeline: Real-time receive sentiment computing engine 0-100 index 2. Member Level Pipeline: Read the level identification of the user account (0-10 levels)

[0343] Core Algorithm

[0344]

[0345]

[0346] 4.10 System Implementation Verification Data

[0347] Test Case 1: Ad Interception during Emotional Low Period

[0348]

[0349] Test Case 2: Premium Ad during High Willingness Period

[0350]

[0351] / / Test Case 3

[0352]

[0353]

[0354] Five, Multi-Patent Coordination Test Data:

[0355]

[0356] Coordination Verification Example

[0357] When implementing this system, the following tests are required to verify the reliability of the full-link degradation:

[0358]

Claims

1. A method for controlling the linkage between emotion and advertising, characterized in that... Includes the following steps: S1: Generate a sentiment index by locally calculating multimodal sensor data on the terminal device; S2: When the sentiment index is below the first threshold, a permission freeze command is sent to the advertising engine; S2.1 When the sentiment index is in the range of 30-70, enable basic advertising permissions (BASIC); S2.2 When the sentiment index is ≥70, enable advanced advertising permissions (PREMIUM); S3: When a preset care event is detected, a linkage operation is executed; S4: Multimodal data is processed at the user terminal, and the raw data does not leave the device; S5: Dynamically adjust the ad slot bidding premium rate based on the sentiment index.

2. An emotion-driven dynamic control system for advertising permissions, characterized in that... Include: a) Edge computing devices process multimodal sensor data in real time; b) Three-level permission state machine (LOCKED / BASIC / PREMIUM); c) A decoupled points-based reward mechanism between care event triggers and the advertising system; d) A level adapter module that maps membership levels to advertising quota multiplier factors; e) A location engine that dynamically adjusts the advertising range through LBS fences; f) A closed-loop optimization pipeline for consumer feedback to the sentiment computing engine; g) Coupon eligibility adjuster: dynamically adjusts the contribution value freeze ratio based on the real-time sentiment index; h) The terminal device includes an emotion computing chip, an advertising permission status indicator, a physical emergency stop button, and an SM4 national cryptographic security module.

3. The system as described in claim 1, characterized in that... Includes a feedback optimization engine, which executes: a) Real-time adjustment of the sentiment index based on consumer satisfaction feedback; b) The updated sentiment index triggers a change in permission status; c) Form a closed loop of "consumption-feedback-optimization".

4. The system as described in claim 1, characterized in that... Execute simultaneously: k) Emotional state → Advertising permission mapping; l) Membership Level → Geofencing Expansion; m) Consumption feedback → Sentiment index correction.

5. The system as described in claim 1, characterized in that... Include: a) A sentiment coefficient labeling output interface provides coefficient values ​​to the financial system; b) Dynamic pricing module for paid advertising, execute: price = min(LTV / 1000,20)*0.5+(EmoScore*TimeFactor)*0.3+max(TimeFactor-1,0)*0.2 in: EmoScore = Mood Index / 100 (normalized to 0-1); TimeFactor is the time-period factor (peak hours = 1.5, off-peak hours = 1.0); The final advertising bid will be subject to a membership level coefficient: final price × (1 + 0.05 × membership level).

6. The system as described in claim 1, characterized in that... Include: a) The bank splitting instruction generator drives the distribution of consumption rebates according to the emotional state multiple by writing the emotion coefficient (emotion_factor) into the extended field of the ISO 20022 message. When the field is missing, it is forcibly overwritten to 1.

0. The splitting message must contain error_code.

7. The system as described in claim 1, wherein the sentiment index is calculated using dynamic weights: Sentiment Index Formula Confidence decay function μ(μ=1 / (1+e -5( δ -0.2 ))) a) The weight value is stored in the device's secure area using SM4 encryption; b) When an external system calls the emotion coefficient interface and the timeout occurs (>500ms), a safety value of 1.0 is returned and a downgraded state is triggered (the error code EMOTION_TIMEOUT is recorded); c) The intent intensity (I) is calculated using the consumer behavior weighting formula. calculate. t represents the time decay of the behavior (unit: minutes). w type Merchant type weights (Food & Beverage = 0.8, Retail = 1.0) The range of W is [0.08, 1.0] (when t = 0, W = w). type (When t→∞, W→0) Based on this, an intent strength score is generated, and the intensity of consumption intent is calculated.

8. The system as described in claim 1, characterized in that... Includes an LBS dynamic engine, execute: a) Preset basic range according to industry type (catering - 3KM, KTV - 8KM, other industries 1-10KM according to category); b) When the sentiment index is <50, the actual range = the base range × 0.7; c) When the emotion index is ≥50, the range is expanded according to the formula "actual range = industry basic range × emotion coefficient + (member level - 2) × 500m". d) Membership level expansion requires an emotional index ≥ 50 (if the emotional index < 50, the membership expansion value will be forcibly reset to zero); e) Bidding is prohibited when the member's sentiment index is less than 30; f) Mandatory lower limit constraint: The actual range shall not be less than 500m; g) The LBS dynamic engine includes an industry basic range preset module, an emotion contraction controller, and a level expansion calculator.

9. The system as described in claim 6, characterized in that... The ISO20022 message extension fields must exist in conjunction with the emotion_factor, geo_radius, and coupon_inflation fields in the associated patent output; implementing any one field alone cannot achieve the complete technical effect.

10. The system as described in claim 1, characterized in that... Includes a "cross-patent collaborative executor" to execute: a) Monitor the sentiment index and permission status output by this system in real time; b) When a critical error (including EMOTION_TIMEOUT) is detected, broadcast the error code to the associated patent system; c) Upon receiving an error notification (such as GEO_FAILURE) from the associated patent system, trigger a local degradation strategy; d) Maintain the error code mapping table: -EMOTION_TIMEOUT → Forces setting emotion_factor = 1.0 -GEO_FAILURE → Switch to BeiDou positioning + accuracy downgrade.

Images