System and method for intelligent placement of user-specific driven advertisements

Abstract

The application belongs to the technical field of advertisement pushing, and discloses an advertisement intelligent putting system and method based on specific user driving: demand expressions voluntarily submitted by users are collected, characteristic elements of each expression item are extracted, and a demand expression characteristic set is obtained; multi-layer condition discrimination is performed on the demand expression characteristic set, and uncoordinated expression items are removed, so that a preliminary putting demand set is obtained; situation derivation enhancement is performed, so that a complete putting demand set is obtained; core expression item data of adjacent demand sequences are extracted, and sequence mapping and fusion processing are performed, so that a persistent preference guidance group is obtained; peripheral expression item data is extracted, and an instant change coordination group is constructed; the peripheral expression items in the complete putting demand set are matched and adjusted, so that an adaptive putting advertisement set is obtained; and the precision marketing efficiency of advertisement intelligent putting is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of advertising delivery technology, and more specifically, to an intelligent advertising delivery system and method driven by specific users. Background Technology

[0002] With the deepening development of the digital economy and the gradual decline of the internet traffic dividend, intelligent advertising delivery technology has become a core infrastructure supporting the digital marketing ecosystem. In a market environment where user attention resources are becoming increasingly scarce, achieving efficient matching of advertising content with users' real needs is of key significance for improving marketing conversion efficiency and optimizing user experience. To this end, programmatic advertising delivery technology based on user profiles and behavior prediction has been widely used. These technologies collect users' historical browsing records, click behavior and consumption trajectory, and combine them with machine learning algorithms to build interest preference models to achieve targeted delivery of advertising content.

[0003] Existing intelligent ad delivery technologies have shortcomings in their user demand capture mechanisms, a problem particularly pronounced in consumer decision-making scenarios where user needs are immediate and context-dependent. Specifically, traditional delivery systems employ a passive inference model based on historical behavioral data, analyzing past user behavior to infer current interests rather than capturing the user's true, immediate needs. This inferential demand capture mechanism leads to multiple dilemmas: historical user behavior data experiences time-sensitive decay, and the correlation between past preferences and current needs weakens over time; user needs are highly dynamic across different life situations, and ad matching based on static profiles cannot effectively respond to this. Scene switching; more importantly, users, as passive recipients of advertisements, lack effective channels to actively express their needs. This results in users with genuine purchase intentions not being able to access relevant advertising resources in a timely manner, while users without purchasing intentions frequently receive irrelevant advertising pushes. The current mainstream approach to solving this problem is to increase the dimensions of data collection and improve the accuracy of algorithm predictions. However, this technical path of replacing genuine expression with more refined inferences not only fails to fundamentally bridge the structural gap between user needs and advertising supply, but also gives rise to problems such as increased privacy compliance risks, escalating consumption of computing resources, and a continuous decline in user trust, ultimately restricting the effectiveness of intelligent advertising delivery in the field of precision marketing.

[0004] In view of this, the present invention proposes an intelligent advertising delivery system and method based on specific user-driven approaches to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a user-driven intelligent advertising delivery method, comprising: Step S1: Collect user-submitted requirement expressions, extract the characteristic elements of each expression item, and obtain the requirement expression characteristic set; Step S2: Perform multi-level condition screening on the set of demand expression characteristics and remove inconsistent expression items to obtain the initial deployment demand set. The multi-level condition screening includes timeliness screening, context screening, intensity screening, relevance screening and completeness screening. Step S3: Perform contextual enhancement on the expression items in the initial demand set to obtain the complete demand set; Step S4: Extract core expression item data of adjacent demand sequences based on the complete demand set, and perform sequence mapping and fusion processing on the core expression item data to obtain the persistent preference guidance group; extract peripheral expression item data based on the complete demand set to construct the real-time change coordination group; Step S5: Based on the persistent preference guidance group and the real-time change coordination group, the peripheral expression items of the complete delivery demand set are matched and adjusted to obtain the adaptive delivery ad set.

[0006] User-driven intelligent ad delivery systems include: Requirement extraction module: Collects user-submitted requirement expressions, extracts characteristic elements of each expression item, and obtains a set of requirement expression characteristics; Demand identification module: Performs multi-level condition identification on the demand expression characteristic set and removes inconsistent expressions to obtain the initial deployment demand set. The multi-level condition identification includes timeliness identification, context identification, intensity identification, relevance identification and completeness identification. Context Enhancement Module: Performs context-derived enhancement on the expression items in the initial deployment demand set to obtain the complete deployment demand set; Preference construction module: Extracts core expression item data of adjacent demand sequences based on the complete demand set, and performs sequence mapping and fusion processing on the core expression item data to obtain the persistent preference guidance group; extracts peripheral expression item data based on the complete demand set to construct the real-time change coordination group; Adaptive Push Module: Based on the persistent preference guidance group and the real-time change coordination group, the module matches and adjusts the peripheral expressions in the complete delivery demand set to obtain an adaptive delivery ad set.

[0007] The technical effects and advantages of the present invention, based on a user-driven intelligent advertising delivery system and method, are as follows: This invention employs multi-level conditional screening—including timeliness, context, intensity, association, and completeness—to identify the characteristic set of demand expressions. It eliminates incompatible expressions to obtain a preliminary demand set, effectively filtering historical preference data with diminishing timeliness and demand expressions that do not match the current context, ensuring the timeliness and contextual adaptability of the delivery demands. User needs in different life scenarios are highly dynamic; a single demand expression cannot cover users' related needs in similar situations. By performing contextual enhancement on the expressions in the preliminary demand set, a complete demand set is obtained, capturing the context-dependent characteristics of user needs and expanding potential needs in relevant contexts, thus compensating for the shortcomings of single demand expressions in contextual coverage. User advertising needs simultaneously contain long-term stable preference characteristics and dynamically changing characteristics; relying solely on a single dimension feature is insufficient. Traditional static ad matching can easily lead to biased targeting. By extracting core expression data from adjacent demand sequences and performing sequence mapping and fusion processing, a persistent preference guidance group is obtained, presenting users' long-term stable interest tendencies. By extracting peripheral expression data, an instant change coordination group is constructed to capture the instant demand change characteristics of users in the current context. Based on the persistent preference guidance group and the instant change coordination group, the peripheral expression items in the complete demand set are matched and adjusted to obtain an adaptive ad set. This allows ad targeting to respond to both users' long-term preference trends and adapt to users' instant demand changes, avoiding the problem that traditional static profile matching cannot effectively respond to scene switching. At the same time, because it is based on users' active expression rather than passive data collection, it reduces privacy compliance risks and computing resource consumption, increases users' trust in the ad targeting system, and thus improves the precision marketing effectiveness of intelligent ad targeting. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the user-driven intelligent advertising delivery method of the present invention; Figure 2 This is a schematic diagram of the user-driven intelligent advertising delivery system of the present invention. Detailed Implementation

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0010] Example 1 Please see Figure 1 As shown, this embodiment of the intelligent ad delivery method driven by specific users includes: Step S1: Collect user-submitted requirement expressions, extract the characteristic elements of each expression item, and obtain the requirement expression characteristic set.

[0011] With the deepening development of the digital economy and the gradual decline of internet traffic dividends, intelligent ad delivery technology has become a core infrastructure supporting the digital marketing ecosystem. Existing intelligent ad delivery technologies employ a passive inference model based on historical behavioral data, analyzing past user behavior to predict current interests rather than capturing users' true, immediate needs. This inferential demand capture mechanism leads to the time-sensitivity decay of historical user behavior data, with the correlation between past preferences and current needs weakening over time. Therefore, this solution collects user-submitted demand expressions and constructs an ad delivery mechanism driven by users' true intentions, fundamentally solving the structural disconnect between user demand and ad supply.

[0012] This solution is applicable to intelligent advertising delivery scenarios in e-commerce platforms, content distribution platforms, social media platforms, and integrated applications. User-submitted demand expressions refer to consumption intentions, product preferences, or service needs that users spontaneously input through the interactive interface, as opposed to data passively acquired by the system through behavioral tracking.

[0013] Preferably, in some possible implementations of the embodiments of the present invention, the specific implementation process of step S1 includes: Step S11: Receive the user's actively inputted requirement expression through the user interaction interface to obtain the original requirement expression.

[0014] User interaction interfaces include, but are not limited to: requirement input panels in mobile applications, preference settings pages on web pages, voice interaction interfaces of intelligent voice assistants, and requirement dialogue windows in instant messaging tools. Users can submit their requirements through various methods such as text input, voice input, image upload, or structured forms.

[0015] For example, a user on an e-commerce platform might enter "I want to buy a smartphone with long battery life and good camera performance, with a budget between 3,000 and 5,000 yuan, and a preference for Huawei or Xiaomi" into the input panel. This input constitutes a basic requirement. Another user might use a voice assistant to express "I'm currently renovating my new house and need to learn about smart home products," which, after voice recognition and conversion, also forms a basic requirement.

[0016] Step S12: Decompose and separate the original requirement expression, extract the expression item identifier, submission time, user identifier, content description and priority mark to obtain the expression structure set.

[0017] Decomposing the original requirement expression involves breaking down the user's complex requirements into independent, processable expression units. The expression item identifier is a unique code assigned to each expression item by the system, generated using the format "user identifier-timestamp-serial number"; the submission time records the precise time the user submitted the requirement, with millisecond-level accuracy; the user identifier is a unique identity code assigned to the user by the platform; the content description is a standardized requirement text processed by natural language processing; and the priority marker reflects the urgency of the user's request, using an integer scale from 1 to 5, where 5 represents the highest priority.

[0018] Taking the aforementioned smartphone demand as an example, after decomposition, three independent expressions are obtained: Expression 1 is described as "a smartphone with long battery life", Expression 2 is described as "a smartphone with good camera performance", and Expression 3 is described as "budget of 3,000 to 5,000 yuan, brand Huawei or Xiaomi". Each expression is associated with the same user identifier and submission time, but is assigned different priority tags based on the semantic analysis results.

[0019] Step S13: Perform element expansion processing on each expression item in the expression structure set based on the user identifier and priority tag to obtain the expression item expansion feature set.

[0020] The purpose of element expansion processing is to enrich the contextual information of each expression item, giving it more complete semantic features and associative attributes. The specific implementation process includes: The initial data of the current expression item is extracted from the expression structure set to obtain the basic data of the expression item. The basic data of the expression item includes five basic fields: expression item identifier, submission time, user identifier, content description, and priority flag.

[0021] The weighted data is obtained by adjusting the weight allocation of the basic data of the expression items based on the priority marker. The calculation method for the weight allocation adjustment is as follows: divide the priority marker value by the highest priority value of 5 to obtain the basic weight coefficient; if the priority marker is 5, the basic weight coefficient is 1.0; if the priority marker is 3, the basic weight coefficient is 0.6. The weighted data adds a weight coefficient field to the basic data of the expression items.

[0022] Based on user identifiers, the weighted adjustment data is extended with historical associations to obtain extended association data. Historical association extension refers to retrieving historical expressions from the user's historical demand database whose content description similarity to the current expression exceeds a preset similarity threshold, and extracting information such as category tags, satisfaction status, and feedback ratings. Text similarity is evaluated based on text fingerprints and fragment matching. The text is cut into fixed-size "fragments" (such as sentences or fixed-length phrases), and a unique "fingerprint" (hash value) is generated for each fragment. Then, the overlap of the fingerprint sets of two texts is compared. In this embodiment, the preset similarity threshold is set to 0.65. When the similarity is below 0.65, the semantic association between the historical expression and the current expression is weak, and introducing too much weak association information will increase noise interference. When the similarity is above 0.65, the historical expression can provide effective contextual supplementation for the current expression.

[0023] The associated extended data is integrated to obtain an extended feature set for expression items. Element integration involves organizing the basic fields and extended fields according to a predefined data structure to form standardized expression item records. A compatibility check is performed on the extended feature set to ensure coordination between elements, resulting in a checked extended feature set for expression items. The compatibility check includes: verifying whether the weight coefficients are within the valid range, verifying the timeliness of historical association information, and confirming the consistency of data types for each field. Supplementary tags are added to the checked extended feature set for expression items to obtain an enhanced extended feature set for expression items. Supplementary tags include demand category tags, consumption scenario tags, and user profile association tags, which are automatically generated by calling the platform's tag classification service.

[0024] Step S14: Extract features from the extended feature set of the expression items to obtain the feature record group of the expression items. The feature record group of the expression items includes the timeliness tag, context description, intensity value, related links and completeness score of each expression item. Construct a standard format requirement expression feature set based on the feature record group of the expression items.

[0025] Feature extraction involves extracting key feature indicators from the enhanced feature set to form standardized feature records. The extraction methods for each feature indicator are as follows: Timeliness Tag: Based on the time difference between the submission time of the expression and the current time, combined with the average validity period of the demand category, the estimated expiration time of the expression is calculated and marked. For example, the average validity period of electronic product demands is 7 days. If a user submits a smartphone purchase demand on October 15, 2023, the timeliness tag will be marked as "October 22, 2023".

[0026] Contextual description: Based on the content description and supplementary tags of the expression items, natural language processing technology is used to extract the user's usage scenario, purchase motivation, and background information of needs. For example, "decorating a new house" is a typical contextual description, reflecting that the user is currently in the home furnishing consumption decision-making cycle.

[0027] Intensity score: Reflects the urgency and certainty of a user's need, ranging from 0 to 1. The intensity score is evaluated by considering priority markers, the degree of certainty in the wording, and the conversion rate of similar historical needs. For example, expressions containing strong words of certainty such as "urgently needed," "immediately," and "must" typically have an intensity score higher than 0.8; while expressions containing weak words of certainty such as "consider," "understand," and "take a look" typically have an intensity score lower than 0.5.

[0028] Related Links: Records the semantic relationships between the current expression and other expressions from the same user, stored in the form of a list of expression identifiers. If a user submits both "smartphone" and "phone accessories" requests, these two expressions are related links to each other.

[0029] Completeness Score: This score assesses the completeness of the information described in the expression item, ranging from 0 to 100. The completeness score is calculated by pre-setting standard information dimensions corresponding to the demand category, calculating the proportion of dimensions covered by the current expression item relative to the total number of standard dimensions, and multiplying this proportion by 100 to obtain the score. For example, the standard information dimensions for smartphone purchase demands include five dimensions: price range, brand preference, core functions, appearance requirements, and purchase timeframe. If a user's expression item covers three dimensions—price range, brand preference, and core functions—then the completeness score is 60. A standardized set of demand expression characteristics is constructed based on the expression item characteristic record group, providing standardized data input for subsequent multi-level conditional identification.

[0030] Step S2: Perform multi-level condition screening on the set of demand expression characteristics and remove inconsistent expression items to obtain the initial deployment demand set. The multi-level condition screening includes timeliness screening, context screening, intensity screening, relevance screening and completeness screening.

[0031] Because user-submitted requests may contain outdated information, vague contexts, weak intent, isolated or irrelevant details, or incomplete information, directly using them for ad targeting will lead to decreased ad performance and a poor user experience. Therefore, it is necessary to conduct multi-level screening of the request expression characteristics to select high-quality, effective request expressions, thereby improving the accuracy and conversion rate of ad targeting.

[0032] Step S21: Based on the configuration of the user interface and the type of requirement, set the validity period range, context matching threshold, intensity lower limit, association intensity threshold and minimum completeness score.

[0033] The criteria for setting each screening threshold are as follows: Validity period range: This range is differentiated based on the consumption decision-making cycle characteristics of different demand types. For fast-moving consumer goods (FMCG), the validity period is set at 1 to 3 days, as users have short purchase decision cycles and their needs change frequently. For electronics, the validity period is set at 7 to 14 days, as users typically require a longer period of information gathering and comparison before purchasing electronics. For major consumer goods, the validity period is set at 30 to 90 days, as the decision-making cycle for major purchases such as real estate and automobiles is longer.

[0034] Context matching threshold: set to 0.55. When the context matching threshold is below 0.55, the context description of the expression item matches the standard context template poorly, making it impossible to accurately determine the user's real consumption scenario, which may lead to the ad content deviating from the user's actual needs; when it is above 0.55, the context description has sufficient clarity to support subsequent ad matching.

[0035] Intensity lower limit: Set to 0.35. An intensity lower limit below 0.35 usually reflects that the user is in the early stage of demand or browsing and understanding, and has not yet formed a clear purchase intention. At this time, the conversion effect of advertising is poor and may cause user aversion. When the intensity value is higher than 0.35, the user has a certain degree of consumption intention, and advertising can effectively promote conversion.

[0036] Association strength threshold: set to 0.40. Association strength reflects the degree of semantic association between an expression item and its adjacent expressions. When it is below 0.40, the expression item exhibits isolated features and it is difficult to infer the user's complete demand graph through contextual association. When it is above 0.40, the expression item can form an effective demand chain with other expressions.

[0037] Minimum completeness score: set at 45 points. A completeness score below 45 points means that the expression covers less than half of the information dimensions, and the lack of key information will increase the ambiguity of ad matching; when the score is above 45 points, the amount of information contained in the expression is sufficient to support basic ad matching needs.

[0038] Step S22: Determine the validity period of each expression item in the demand expression characteristic set, and remove expression items whose validity period tags exceed the validity period range to obtain the validity period discrimination set.

[0039] The specific method for determining the validity period is as follows: Obtain the current system time and compare it with the validity period tag of each expression item. If the validity period tag is earlier than the current time, it is determined to be expired. At the same time, check whether the time difference between the submission time and the current time exceeds the upper limit of the validity period range corresponding to the requirement type. If it exceeds, it is also determined to be expired. Expired expression items are removed from the requirement expression characteristic set, and the remaining expression items constitute the validity period screening set.

[0040] Step S23: For each expression item in the demand expression characteristic set, determine the context matching degree and remove expression items whose context description is lower than the context matching degree threshold to obtain the context discrimination set.

[0041] The context matching score is calculated as follows: the context description of an expression item is compared with a pre-defined standard context template library for semantic similarity calculation. The highest similarity value is taken as the context matching score of that expression item. It should be noted that common algorithms such as word overlap or TF-IDF are used for semantic similarity calculation. The standard context template library contains typical descriptions of various consumption scenarios, such as "daily shopping," "holiday gift-giving," "home decoration," and "business travel." If the context matching score of an expression item is lower than 0.55, its context description is determined to be vague or incomplete, and it is removed from the set of demand expression characteristics. The remaining expression items constitute the context discrimination set.

[0042] Step S24: Calculate the description deviation of each expression item in the demand expression characteristic set to the core expression item, remove expression items whose deviation exceeds the intensity lower limit, and obtain the intensity discrimination set.

[0043] The core expression item refers to the expression item with the highest intensity value among the demand expressions submitted by the same user in the same batch, representing the user's most urgent needs and desires. The description deviation is calculated as follows: obtain the semantic vectors of the content descriptions of the current expression item and the core expression item, and calculate the cosine distance between the two vectors as the description deviation. If the description deviation exceeds the lower limit of intensity of 0.35 and the intensity value of the current expression item itself is lower than 0.35, then the expression item is determined to be a marginal demand and is removed from the demand expression characteristic set. The remaining expression items constitute the intensity discrimination set.

[0044] Step S25: Calculate the link strength between each expression item in the demand expression characteristic set and its neighboring expression items, and remove expression items whose strength is lower than the association strength threshold to obtain the association discrimination set.

[0045] Adjacent expressions refer to expressions submitted by the same user that are adjacent in time or semantically related. Link strength is calculated by considering both temporal proximity and semantic relevance. Temporal proximity is represented by the reciprocal of the difference in submission times, while semantic relevance is represented by the cosine similarity of the content descriptions. Link strength is the weighted sum of temporal proximity and semantic relevance, with weighting coefficients of 0.3 and 0.7, respectively. If the link strength of an expression is below 0.40, it is considered an isolated request and removed from the set of request expression characteristics. The remaining expressions constitute the association screening set.

[0046] Step S26: Perform a completeness score check on each expression item in the demand expression characteristic set, and remove expression items with scores lower than the lowest completeness score to obtain the completeness screening set.

[0047] The completeness score check directly compares the completeness score of the expression item with the lowest completeness score of 45. Expression items with scores below 45 are removed from the demand expression characteristics set because they are severely lacking in information and cannot effectively support ad matching. The remaining expression items constitute the completeness screening set.

[0048] Step S27: Integrate the timeliness discrimination set, context discrimination set, intensity discrimination set, association discrimination set, and integrity discrimination set into multiple layers, retain the expression items that meet all discrimination conditions, and obtain the preliminary deployment demand set.

[0049] The multi-layered integration employs an intersection operation: First, a set of expression item identifiers from five screening sets is obtained. An intersection operation is then performed on these five sets to obtain the expression item identifiers that exist simultaneously in all screening sets. Based on this identifier set, corresponding expression items are extracted from the demand expression characteristic set to form a preliminary demand set. Through multi-level conditional screening, low-quality demand expression items are effectively filtered out, ensuring that expression items entering subsequent processing stages all possess the characteristics of timeliness, clear context, sufficient intensity, close relevance, and complete information.

[0050] Step S3: Perform contextual enhancement on the expression items in the initial deployment demand set to obtain the complete deployment demand set.

[0051] Users' needs are highly dynamic across different life situations, and the needs they actively submit often only reflect their explicit needs, ignoring implicit and derivative needs associated with those explicit needs. For example, when a user expresses a need to "buy a smartphone," they may also have derivative needs such as "buy a phone case" or "select wireless headphones." If advertising is based solely on explicit needs, related consumption opportunities will be missed, reducing the completeness of user needs being met. Therefore, it is necessary to perform contextual enhancement on the initial demand expressions to uncover and supplement users' implicit and derivative needs.

[0052] Step S31: Based on the input frequency and expression structure information of the user interface, perform contextual branch recognition on the expression items in the initial delivery demand set to obtain the expression item group to be enhanced.

[0053] Input frequency reflects the density of user requests submitted per unit of time, while expression structure information reflects the syntactic structure and semantic hierarchy of the requests. The purpose of contextual branch identification is to filter out expressions with derivation potential.

[0054] The rules for identifying contextual branches include: if the content description of an expression contains a parallel structure (such as "A and B", "A and B"), then a contextual branch is determined to exist; if the standard context template corresponding to the context description of an expression has a preset chain of derived contexts, then a contextual branch is determined to exist; if the same user submits multiple related expressions at a high frequency within a short period of time (input frequency exceeding 2 times per minute), then these expressions may have unexpressed related contextual branches. Expressions that meet any of the above conditions are included in the group of expressions to be enhanced.

[0055] Step S32: Perform context-derived matching calculation for each expression item in the group of expression items to be enhanced, obtain the enhancement factor, and obtain the enhancement parameter group.

[0056] The specific process for calculating the situation-derived proportions is as follows: For each expression in the enhanced expression group, extract description information and corresponding submission time to construct an expression description record, resulting in a description record set. The description record contains five fields: expression identifier, original content description, standardized content description, submission time, and the context category.

[0057] For each expression in the description record set, the relative contextual position of the target enhancement item is extracted, and its derived matching factor is calculated to obtain the factor dataset. The target enhancement item refers to the derived demand item associated with the current expression item in the pre-defined contextual derived knowledge base. The relative contextual position reflects the distance between the derived demand and the original demand in the contextual chain; the closer the distance, the larger the derived matching factor. The formula for calculating the derived matching factor is: ;in, For derived proportioning factors; The base derivative coefficient corresponding to the scenario category has a value range of 0.5 to 1.5, with 0.8 for fast-moving consumer goods, 1.0 for electronic products, and 1.2 for bulk consumer goods. The relative contextual position, with a value of a positive integer, directly derives from the requirements. A value of 1 indicates secondary derived demand. The value is 2, and so on; The semantic relevance is denoted by 0, and its value ranges from 0 to 1, reflecting the degree of semantic relevance between the derived requirement and the original requirement.

[0058] The factor dataset and description record set are associated and paired, and descriptive context derivation is performed to obtain a derived description set. Descriptive context derivation involves extracting derived requirement descriptions from the context derivation knowledge base based on the derivation matching factor and performing semantic fusion with the content descriptions of the original expression items. Submission time and priority marker information are recorded synchronously for both the derived description set and the factor dataset to construct an enhanced parameter set. The submission time of the derived requirement inherits the submission time of the original expression item, and the priority marker is scaled and adjusted according to the derivation matching factor.

[0059] Step S33: Perform description derivation for each enhancement target item in the enhancement parameter group, construct an extended description record, and obtain the enhancement requirement set.

[0060] The derivation process is as follows: Derived descriptions and derived matching factors for each enhancement target item are extracted from the enhancement parameter set. The derived descriptions are then processed using the method in step S14 to generate timeliness tags, contextual descriptions, intensity values, associated links, and completeness scores. The intensity value is determined by the product of the original expression item's intensity value and the derived matching factor, and the associated links point to the original expression item's identifier. The enhancement target items after feature refinement constitute the enhancement requirement set.

[0061] Step S34: Combine the initial deployment demand set and the enhanced demand set to construct a complete deployment demand set.

[0062] The method for joint splicing is as follows: Maintain the integrity of the original expressions in the initial demand set, and insert the derived expressions from the enhanced demand set after the corresponding original expressions according to the pointing relationship of the links, forming a chain structure of "original demand - derived demand". If the same original expression corresponds to multiple derived expressions, they are arranged in descending order of the derived matching factor.

[0063] Step S35: Perform derivative coherence verification on the complete deployment requirement set to ensure the consistency of the description of the enhancements, and obtain the verified complete deployment requirement set.

[0064] The purpose of derivation coherence verification is to examine the semantic consistency between derived expressions and original expressions, avoiding the introduction of irrelevant derived requirements due to the limitations of the derived knowledge base. The verification method is as follows: calculate the semantic similarity between the content description of the derived expression and the original expression. If the similarity is below 0.50, the derived expression is deemed inconsistent with the original requirement and removed from the complete delivery requirement set. Simultaneously, it is checked whether the contextual description of the derived expression is on the same contextual chain as the contextual description of the original expression; if not, it is also removed. Through contextual derivation enhancement, the complete delivery requirement set not only includes the user's explicitly expressed requirements but also covers derived requirements associated with explicit requirements, providing users with a more comprehensive advertising recommendation service and improving the completeness of user need fulfillment and the coverage efficiency of advertising delivery.

[0065] Step S4: Extract core expression data of adjacent demand sequences based on the complete demand set, and perform sequence mapping and fusion processing on the core expression data to obtain the persistent preference guidance group; extract peripheral expression data based on the complete demand set to construct the real-time change coordination group.

[0066] In traditional advertising models, users, as passive recipients of ads, lack effective channels for actively expressing their needs, preventing genuinely interested buyers from accessing relevant advertising resources in a timely manner. This step addresses this by differentiating between core and peripheral expressions, constructing two complementary demand representation structures: a persistent preference guidance group and an immediate change coordination group. This achieves a two-dimensional characterization of both stable and dynamic user preferences, thereby improving the accuracy of matching ad placement with users' actual needs.

[0067] Preferably, in some possible implementations of the embodiments of the present invention, the specific process of extracting core expression item data of adjacent demand sequences based on the complete demand set, and performing sequence mapping and fusion processing on the core expression item data includes: The complete set of delivery requests is arranged chronologically according to submission time, dividing it into multiple adjacent request sequence windows. Since user requests submitted at different times may reflect their consumption intentions at different stages, mixing all requests together would make it difficult to accurately capture the temporal evolution of user preferences. The specific implementation process for chronological arrangement is as follows: The submission time field of all requests in the complete set of delivery requests is obtained, and the requests are rearranged using an ascending sorting algorithm, placing the request with the earliest submission time at the beginning of the sequence.

[0068] The partitioning of adjacent demand sequences employs a sliding time window mechanism, with a window size of 24 hours and a sliding step size of 12 hours. The 24-hour window size is based on the fact that users' daily consumption decisions are typically made on a daily cycle, and 24 hours can cover the user's expressed needs within a complete daily cycle. A window size that is too small would result in insufficient items expressed within a single window, while a window size that is too large would confuse the intent of demands across different time periods. Setting the sliding step size to half the window size ensures overlap between adjacent windows, preventing the artificial segmentation of continuous demand sequences due to window boundary cutting.

[0069] Calculate the content consistency weight for each expression item within each adjacent demand sequence window, identify expression items with a content consistency weight greater than a preset core weight threshold as core expression items, and construct a core expression item set.

[0070] Content consistency weight measures the semantic consistency between a single expression and other expressions within the same window, reflecting the representativeness of that expression in the current demand sequence. The calculation process includes: converting each content description into a semantic vector using a predefined semantic encoding model (e.g., Word2Vec, FastText); calculating the cosine similarity between the semantic vector of each expression within the window and the semantic vectors of all other expressions within the window; averaging the similarity values ​​and determining the content consistency weight based on the mean, using the following formula: ; in, Weighting for content consistency; This represents the number of other items displayed in the window. For the current expression item and the first Cosine similarity between other expressions; For the first The intensity value of each of the other expressions.

[0071] The semantic association strength value is evaluated for any two core expressions in the core expression set. Core expression pairs with semantic association strength values ​​greater than a preset association threshold are included in the same mapping group. The preset core weight threshold is set to 0.60, with a value range of 0.50 to 0.75. When the threshold is below 0.50, many expressions with weak association with the window topic will be misjudged as core expressions; when the threshold is above 0.75, important user needs and intentions may be missed.

[0072] The semantic association strength value is calculated by weighting and summing three dimensions: semantic similarity, temporal proximity, and attribute consistency, with weights of 0.50, 0.25, and 0.25, respectively. The formula for calculating temporal proximity is: ; in, Represents temporal proximity. Represents a time interval. The basic time span is set to 48 hours in this embodiment; attribute consistency is obtained by comparing the matching degree of the demand category tag and consumption scenario tag of the two core expression items. The value is 1 when there is a complete match, 0.5 when there is a partial match, and 0 when there is no match.

[0073] The preset association threshold is set to 0.55, with a range of 0.45 to 0.70. When the threshold is below 0.45, core expression pairs with weak semantic associations will be incorrectly grouped into the same group; when the threshold is above 0.70, it may lead to too many isolated groups.

[0074] Set fusion trigger conditions for each mapping group. The fusion trigger conditions include: the number of expression items within the group is greater than a preset number threshold and the semantic distance between groups is less than a preset distance threshold.

[0075] The preset threshold for the number of expressions is set to 2, with a range of 2 to 5. Groups consisting of a single expression are insufficient to fully reflect stable user preferences; groups containing two or more expressions indicate that the preference topic has high fusion value. The semantic distance between groups is calculated using the Euclidean distance between the centroid vectors of the two groups. The preset distance threshold is set to 0.80, with a range of 0.60 to 1.00. When the semantic distance is greater than 0.80, the preference topics represented by the two groups differ significantly, and forced fusion will lead to ambiguity in preference orientation.

[0076] For mapping groups that meet the fusion triggering conditions, perform iterative fusion processing. In each iteration, select the two groups with the smallest semantic distance to merge, and terminate the iteration when no group meets the fusion triggering conditions.

[0077] The iterative fusion process employs a hierarchical clustering approach, with the following steps: First, obtain all mapping groups that satisfy the fusion triggering conditions and calculate the semantic distance between any two groups. Second, select the pair of groups with the smallest semantic distance and merge them; the new group contains all the core representations of the original two groups. Third, recalculate the semantic distance between the new group and other groups, updating the fusion triggering condition's determination status. Fourth, check if there are any groups that satisfy the conditions; if so, continue iterating; otherwise, terminate. The centroid vector of the merged new group is calculated using the weighted average of the semantic vectors of each core representation within the group.

[0078] The merged groups are evaluated for preference persistence scores. Groups with preference persistence scores greater than a preset persistence threshold are output as persistent preference guidance groups. The preference persistence score comprehensively considers three factors: time span, expression frequency, and content stability. Each factor is weighted and merged based on maximum-minimum normalization to obtain the preference persistence score. The weight coefficients are set to 0.40, 0.30, and 0.30, respectively. The preset persistence threshold is set to 0.50, with a value range of 0.40 to 0.65. Groups with scores below 0.50 may only reflect short-term, temporary user needs and lack the stability to serve as persistent preference guidance. Through the above process, a complete workflow is achieved, accurately identifying core expression items from a complete set of delivery needs, effectively integrating preference information across time periods, and selecting stable and persistent preferences.

[0079] Preferably, in some possible implementations of the embodiments of the present invention, the specific process of extracting peripheral expression item data based on the complete deployment demand set and constructing an instant change coordination group includes: Extract peripheral expression items from adjacent demand sequences within the complete demand set to construct a peripheral expression item set. The extraction rule for peripheral expression items is as follows: select all expression items in each demand sequence except for the core expression items, as well as derived expression items added through contextual enhancement.

[0080] The peripheral expression item set is transformed into a region sequence group. The region sequence transformation uses the same mapping transformation method as the core expression item, but with the core expression item as the reference point, the peripheral expression items are arranged according to their association distance with the core expression item.

[0081] Based on the peripheral expression item region sequence group, hierarchical deviation fusion is used to perform P-level processing to construct an instant change coordination group. P-level processing is a hierarchical deviation adjustment method, where P is 3, indicating that peripheral expressions are divided into 3 levels according to their degree of deviation from the core expression item: Level 1 consists of peripheral expressions with the smallest deviation, which are highly correlated with the core expression item but have slight differences in some dimensions; Level 2 consists of peripheral expressions with a moderate degree of deviation, which are significantly different from the core expression item but still fall within the same demand category; Level 3 consists of peripheral expressions with the largest degree of deviation, which may reflect the user's temporary interests or exploratory needs.

[0082] The hierarchical deviation fusion method is as follows: Deviation feature vectors are calculated for each level of peripheral expression item, recording the dimension and magnitude of their difference from the core expression item; the deviation feature vectors of each level are organized into a hierarchical structure according to the level order, forming an instant change coordination group. The instant change coordination group reflects the instantaneous fluctuation characteristics of user needs based on core preferences and is used for subsequent matching and adjustment of peripheral expression items.

[0083] Step S5: Based on the persistent preference guidance group and the real-time change coordination group, the peripheral expression items of the complete delivery demand set are matched and adjusted to obtain the adaptive delivery ad set.

[0084] Step S51: Extract the content description vector and basic weight coefficients of each peripheral expression item to construct the feature matrix of the peripheral expression item.

[0085] The content description vector is extracted through a pre-trained semantic encoding model, and the basic weight coefficients are the weights assigned based on priority labels in step S13. The feature matrix of peripheral representation items is constructed by concatenating the content description vector of each peripheral representation item with its priority weight and stacking them in chronological order.

[0086] Step S52: Calculate the instantaneous deviation value for each peripheral expression item in the feature matrix of peripheral expression items based on the instantaneous change coordination group, calculate the persistent matching value for each peripheral expression item based on the persistent preference guidance group, and weight and combine the instantaneous deviation value and the persistent matching value to obtain the comprehensive adjustment index.

[0087] Because user needs possess both immediate dynamism and long-term stability, relying solely on immediate changes or persistent preferences for adjustments is one-sided. This step calculates immediate deviation values ​​and persistent matching values ​​separately, and then performs a weighted combination to achieve a comprehensive assessment of the necessity for adjusting peripheral expression terms.

[0088] The formula for calculating the instantaneous deviation value is: ; in, This is the instantaneous deviation value, ranging from 0 to 1; This is a content description vector for the outermost representation item; The centroid vector of the corresponding level in the instantaneous change coordination group; The maximum normalized distance constant is set to 2.0.

[0089] The formula for calculating persistent match values ​​is: ; in, For persistent matching values; This is a persistent preference feature vector.

[0090] The formula for calculating the comprehensive adjustment index is as follows: ; in, To comprehensively adjust the index; The instantaneous deviation weighting coefficient is set to 0.55; The weighting coefficient for persistent matching is set to 0.45. Immediate changes reflect the user's current dynamic needs and have high timeliness value in advertising scenarios. This value allows immediate changes to be slightly dominant while maintaining sufficient consideration for persistent preferences.

[0091] Step S53: Set the adjustment start threshold and adjustment magnitude threshold, identify peripheral expression items whose comprehensive adjustment index is greater than the adjustment start threshold as expression items to be adjusted, and construct a queue of expression items to be adjusted.

[0092] The adjustment initiation threshold is set to 0.40, with a range of 0.30 to 0.55. When the overall adjustment index is below 0.40, it indicates that the overall deviation of the peripheral expressions is small, and maintaining the original state is sufficient to meet the ad matching requirements; when it is above 0.40, it indicates that there is a clear need for adjustment. The adjustment magnitude threshold is set to 0.25, with a range of 0.15 to 0.40, and is used to determine whether the adjustment is sufficient during subsequent iterative adjustments.

[0093] The method for constructing the queue of expression items to be adjusted is as follows: peripheral expression items with a comprehensive adjustment index greater than the adjustment initiation threshold are inserted into the queue in descending order of their comprehensive adjustment index; peripheral expression items with a comprehensive adjustment index less than or equal to the threshold are kept in their original state.

[0094] Step S54: Perform cyclic adjustment processing on the queue of expression items to be adjusted. In each cycle, select the expression item with the largest comprehensive adjustment index for priority adjustment, update its content description vector and recalculate the coordination degree. The cycle ends when the comprehensive adjustment index of all expression items to be adjusted in the queue is greater than the adjustment magnitude threshold and the coordination degree with adjacent expression items is lower than the preset coordination threshold, thus obtaining the set of outer expression items to be adjusted.

[0095] Traditional adjustment methods employ independent adjustment strategies, neglecting the coordination relationship between adjusted expressions and adjacent expressions. This step introduces a coordination mechanism based on bidirectional association evaluation to ensure that the adjustment result of each expression maintains semantic coherence with its preceding and following adjacent expressions.

[0096] The default coordination threshold is set to 0.50, with a range of 0.40 to 0.65. A coordination score higher than 0.50 indicates sufficient semantic association between the current expression and its adjacent expressions; a score lower than 0.50 indicates insufficient semantic association, requiring further adjustment.

[0097] The specific process of cyclic adjustment includes: selecting the expression term with the largest comprehensive adjustment index from the queue to be adjusted as the current adjustment target; calculating the adjustment direction vector. ,in, The directional mixing coefficient is set to 0.60; the adjustment step size is calculated. ,in The step scaling factor is set to 0.30, with a range of 0.20 to 0.50; Update the content description vector: ; in, This represents the updated vector. A unit vector representing the persistent preference feature vector.

[0098] The degree of coordination is calculated using a two-way correlation assessment method: the content correlation score between the current expression and the preceding expression is calculated separately. The content-related score of the expression item. The content association score is calculated using cosine similarity based on the word frequency vectors obtained from word frequency statistics; the harmonic mean of the two association scores is taken as the degree of reconciliation. ; in, This represents the degree of harmony. The harmonic mean is used because it is more sensitive to smaller values, ensuring that the adjusted expression maintains a good correlation with its preceding and following expressions.

[0099] The loop termination condition is: the overall adjustment index of all expressions to be adjusted in the queue is less than or equal to the adjustment magnitude threshold of 0.25, and the coordination degree with adjacent expressions is greater than or equal to the preset coordination threshold of 0.50. The maximum number of loops is set to 50 to avoid infinite loops.

[0100] Step S55: Merge the adjusted peripheral expression item set with the unadjusted core and internal expression items in the complete delivery requirement set, reconstruct the expression item structure of all sequences, and obtain the adaptive delivery ad set.

[0101] The merging and reconstructing process includes: maintaining the original structure of the core and internal expressions unchanged, with the core expressions serving as the main anchor of the demand sequence; reordering the adjusted peripheral expressions according to their cosine similarity with the core expressions from high to low; inserting peripheral expressions not included in the adjustment into the sequence structure according to their original positional relationships; and performing structural integrity checks, including verifying the validity of associated links, the correctness of the temporal structure, and whether the semantic similarity of adjacent expressions is within a reasonable range.

[0102] The adaptive ad set includes core expressions, adaptively adjusted peripheral expressions, internal expressions, and unadjusted peripheral expressions with low deviation. When matching ads based on this ad set, the system prioritizes matching ad resources corresponding to the core expressions, while expanding the recommendation range based on the characteristics of the adjusted peripheral expressions, achieving accurate matching and dynamic adaptation of ad content with users' multi-level needs.

[0103] This solution establishes a channel for users to proactively express their needs, enabling users with genuine purchasing intentions to access relevant advertising resources in a timely manner. At the same time, it avoids pushing irrelevant advertising content to users without purchasing intentions. This fundamentally solves the structural disconnect between user demand and advertising supply caused by existing inferential demand capture mechanisms. While improving the accuracy of advertising, it also protects user privacy, reduces computing resource consumption, and effectively enhances the overall efficiency of intelligent advertising in the field of precision marketing.

[0104] Example 2

[0105] Please see Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. A user-driven intelligent advertising delivery system is provided, including: Requirement extraction module: Collects user-submitted requirement expressions, extracts characteristic elements of each expression item, and obtains a set of requirement expression characteristics; Demand identification module: Performs multi-level condition identification on the demand expression characteristic set and removes inconsistent expressions to obtain the initial deployment demand set. The multi-level condition identification includes timeliness identification, context identification, intensity identification, relevance identification and completeness identification. Context Enhancement Module: Performs context-derived enhancement on the expression items in the initial deployment demand set to obtain the complete deployment demand set; Preference construction module: Extracts core expression item data of adjacent demand sequences based on the complete demand set, and performs sequence mapping and fusion processing on the core expression item data to obtain the persistent preference guidance group; extracts peripheral expression item data based on the complete demand set to construct the real-time change coordination group; Adaptive Push Module: Based on the persistent preference guidance group and the real-time change coordination group, the module matches and adjusts the peripheral expressions in the complete delivery demand set to obtain an adaptive delivery ad set.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A user-driven intelligent ad delivery method, characterized in that, include: Step S1: Collect user-submitted requirement expressions, extract the characteristic elements of each expression item, and obtain the requirement expression characteristic set; Step S2: Perform multi-level condition screening on the set of demand expression characteristics and remove inconsistent expression items to obtain the initial deployment demand set. The multi-level condition screening includes timeliness screening, context screening, intensity screening, relevance screening and completeness screening. Step S3: Perform contextual enhancement on the expression items in the initial demand set to obtain the complete demand set; Step S4: Extract the core expression item data of adjacent demand sequences based on the complete demand set, and perform sequence mapping and fusion processing on the core expression item data to obtain the persistent preference guidance group; Extract peripheral expression data based on the complete deployment demand set, and build an instant change coordination group; Step S5: Based on the persistent preference guidance group and the real-time change coordination group, the peripheral expression items of the complete delivery demand set are matched and adjusted to obtain the adaptive delivery ad set.

2. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Step S1 includes: Step S11: Receive the user's actively inputted requirement expression through the user interaction interface to obtain the original requirement expression; Step S12: Decompose and separate the original requirement expression, extract the expression item identifier, submission time, user identifier, content description and priority mark to obtain the expression structure set; Step S13: Perform element expansion processing on each expression item in the expression structure set based on user identifier and priority tag to obtain the expression item expansion feature set; Step S14: Extract features from the extended feature set of the expression items to obtain the feature record group of the expression items. The feature record group of the expression items includes the timeliness tag, context description, intensity value, related links and completeness score of each expression item. Construct a standard format requirement expression feature set based on the feature record group of the expression items.

3. The intelligent advertising delivery method based on specific user-driven principles according to claim 2, characterized in that, Based on user identifiers and priority tags, each expression item in the expression structure set undergoes element expansion processing to obtain an extended feature set for each expression item, including: The initial data of the current expression item is extracted from the expression structure set to obtain the basic data of the expression item. The weight allocation of the basic data of the expression item is adjusted based on the priority label to obtain the weight adjustment data. The weight adjustment data is extended by historical association according to the user identifier to obtain the association extended data. The elements of the association extended data are integrated to obtain the expression item extended feature set. The compatibility check of the expression item extended feature set is performed to ensure the coordination between elements to obtain the checked expression item extended feature set. Supplementary tags are added based on the checked expression item extended feature set to obtain the enhanced expression item extended feature set.

4. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Step S2 includes: Step S21: Based on the configuration of the user interface and the type of requirement, set the validity period range, context matching threshold, intensity lower limit, association intensity threshold and minimum completeness score; Step S22: Determine the validity period of each expression item in the demand expression characteristic set, and remove expression items whose validity period tags exceed the validity period range to obtain the validity period discrimination set; Step S23: For each expression item in the demand expression characteristic set, determine the context matching degree and remove expression items whose context description is lower than the context matching degree threshold to obtain the context discrimination set; Step S24: Calculate the description deviation from the core expression item for each expression item in the demand expression characteristic set, and remove expression items whose deviation exceeds the intensity lower limit to obtain the intensity discrimination set; Step S25: Calculate the link strength between each expression item in the demand expression characteristic set and its neighboring expression items, and remove expression items whose strength is lower than the association strength threshold to obtain the association discrimination set; Step S26: Perform a completeness score check on each expression item in the demand expression characteristic set, and remove expression items with scores lower than the lowest completeness score to obtain the completeness screening set; Step S27: Integrate the timeliness discrimination set, context discrimination set, intensity discrimination set, association discrimination set, and integrity discrimination set into multiple layers, retain the expression items that meet all discrimination conditions, and obtain the preliminary deployment demand set.

5. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Step S3 includes: Step S31: Based on the input frequency and expression structure information of the user interface, perform contextual branch recognition on the expression items in the initial delivery demand set to obtain the expression item group to be enhanced; Step S32: Perform context-derived matching calculations for each expression item in the group of expression items to be enhanced, obtain the enhancement factor, and obtain the enhancement parameter group; Step S33: Perform description derivation for each enhancement target item in the enhancement parameter group, construct an expanded description record, and obtain the enhancement requirement set; Step S34: Combine the initial deployment demand set and the enhanced demand set to construct a complete deployment demand set; Step S35: Perform derivative coherence verification on the complete deployment requirement set to ensure the consistency of the description of the enhancements, and obtain the verified complete deployment requirement set.

6. The intelligent advertising delivery method based on specific user-driven principles according to claim 5, characterized in that, For each expression term in the enhanced expression term group, perform context-derived matching calculations to obtain the enhancement factor, resulting in an enhanced parameter set, including: For each expression in the expression group to be enhanced, extract description information and corresponding submission time to construct expression description records and obtain a description record set; for each expression in the description record set, extract the relative context position of the target enhancement item, calculate its derived matching factor, and obtain a factor dataset; associate and pair the factor dataset and the description record set, perform description context derivation, and obtain a derived description set; synchronously record the submission time and priority label information for the derived description set and the factor dataset to construct an enhancement parameter group.

7. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Based on the complete demand set, core expression item data of adjacent demand sequences are extracted, and sequence mapping and fusion processing are performed on the core expression item data, including: The complete set of delivery requirements is arranged chronologically according to the submission time and divided into multiple adjacent requirement sequence windows. The content consistency weight is calculated for the expression items in each adjacent requirement sequence window. Expression items with a content consistency weight greater than a preset core weight threshold are identified as core expression items, and a core expression item set is constructed. The semantic association strength value is evaluated for any two core expression items in the core expression item set. Core expression item pairs with a semantic association strength value greater than a preset association threshold are included in the same mapping group. Set fusion trigger conditions for each mapping group. The fusion trigger conditions include: the number of expression terms in the group is greater than a preset number threshold and the semantic distance between the groups is less than a preset distance threshold. Perform iterative fusion processing on the mapping groups that meet the fusion trigger conditions. In each iteration, select the two groups with the smallest semantic distance to merge. The iteration ends when no group meets the fusion trigger conditions. The preference persistence score is evaluated for the merged groups, and the groups with preference persistence scores greater than the preset persistence threshold are output as persistent preference guidance groups.

8. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Based on the complete deployment demand set, peripheral expression item data is extracted, and an instant change coordination group is constructed, including: Extract peripheral expression items from adjacent demand sequences in the complete demand set to construct a peripheral expression item set; perform regional sequence transformation on the peripheral expression item set to construct a peripheral expression item regional sequence group; and perform P-level processing using hierarchical deviation fusion based on the peripheral expression item regional sequence group to construct an instant change coordination group.

9. The intelligent advertising delivery method based on specific user-driven principles according to claim 1, characterized in that, Step S5 includes: Step S51: Extract the content description vector and basic weight coefficients of each peripheral expression term to construct the feature matrix of the peripheral expression term; Step S52: Calculate the instantaneous deviation value for each peripheral expression item in the feature matrix of peripheral expression items based on the instantaneous change coordination group, calculate the persistent matching value for each peripheral expression item based on the persistent preference guidance group, and weight and combine the instantaneous deviation value and the persistent matching value to obtain the comprehensive adjustment index; Step S53: Set the adjustment start threshold and adjustment magnitude threshold, identify peripheral expression items whose comprehensive adjustment index is greater than the adjustment start threshold as expression items to be adjusted, and construct a queue of expression items to be adjusted; Step S54: Perform cyclic adjustment processing on the queue of expression items to be adjusted. In each cycle, select the expression item with the largest comprehensive adjustment index for priority adjustment, update its content description vector and recalculate the coordination degree. The cycle ends when the comprehensive adjustment index of all expression items to be adjusted in the queue is greater than the adjustment magnitude threshold and the coordination degree with adjacent expression items is lower than the preset coordination threshold. The set of outer expression items to be adjusted is obtained. The coordination degree with adjacent expression items is based on bidirectional association evaluation. For the current expression item to be adjusted, calculate its content association score with the preceding and following expression items respectively, and take the harmonic mean of the two association scores as the coordination degree. When the coordination degree is lower than the preset coordination threshold, it means that the expression item needs to be adjusted. Step S55: Merge the adjusted peripheral expression item set with the unadjusted core and internal expression items in the complete delivery requirement set, reconstruct the expression item structure of all sequences, and obtain the adaptive delivery ad set.

10. A user-driven intelligent advertising delivery system, used to implement the user-driven intelligent advertising delivery method according to any one of claims 1 to 9, characterized in that, include: Requirement extraction module: Collects user-submitted requirement expressions, extracts characteristic elements of each expression item, and obtains a set of requirement expression characteristics; Demand identification module: Performs multi-level condition identification on the demand expression characteristic set and removes inconsistent expressions to obtain the initial deployment demand set. The multi-level condition identification includes timeliness identification, context identification, intensity identification, relevance identification and completeness identification. Context Enhancement Module: Performs context-derived enhancement on the expression items in the initial deployment demand set to obtain the complete deployment demand set; Preference construction module: Extracts core expression item data of adjacent demand sequences based on the complete demand set, and performs sequence mapping and fusion processing on the core expression item data to obtain persistent preference guidance groups; Extract peripheral expression data based on the complete deployment demand set, and build an instant change coordination group; Adaptive Push Module: Based on the persistent preference guidance group and the real-time change coordination group, the module matches and adjusts the peripheral expressions in the complete delivery demand set to obtain an adaptive delivery ad set.

Images