Object evaluation method and apparatus

Abstract

This application provides an object evaluation method and apparatus. The object evaluation method includes: acquiring object parameters of the target object to be evaluated in the ontology attribute dimension and the interaction attribute dimension; based on the object parameters in the ontology attribute dimension and the interaction attribute dimension, obtaining the target ontology feature vector in the ontology attribute dimension and the target interaction feature vector in the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model, wherein the feature analysis layer is determined based on the attributes of the object parameters in the corresponding attribute dimensions; and fusing the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object. This solution extracts ontology and interaction attribute dimension parameters and extracts features specifically through an adapted feature analysis layer, improving the comprehensiveness and accuracy of value evaluation. It can be applied to digital cultural product production software in the field of next-generation information technology (such as virtual reality).

Description

Technical Field

[0001] This application relates to the field of next-generation information technology, and in particular to an object evaluation method. This application also relates to an object evaluation apparatus, a computing device, a computer-readable storage medium, and a computer program. Background Technology

[0002] Massively multiplayer online role-playing games (MMORPGs) rely heavily on player communities and feature group gameplay such as team play, guilds, and faction wars. They build a sense of long-term companionship through continuous social interaction among players, thereby increasing user stickiness.

[0003] However, these types of games are complex, and new players often face difficulties integrating and lack of guidance, leading to high early churn rates. If experienced core players act as mentors to guide newcomers and help them integrate into the community, new player retention rates can be significantly improved. Furthermore, MMORPGs exhibit community-based monetization, with core players acting as trendsetters, influencing other players to follow suit. Core players can also boost team win rates through their excellent leadership skills, effectively driving overall player engagement.

[0004] Currently, the primary method for selecting core players is through statistical analysis of data such as player combat power, spending, and level. However, core players selected using this method often fail to effectively activate community activity or boost the overall activity of both new and existing players, making it difficult to truly leverage their positive driving force on the game ecosystem. Therefore, a method that can accurately select high-value players is urgently needed. Summary of the Invention

[0005] In view of this, embodiments of this specification provide an object evaluation method. One or more embodiments of this specification also relate to an object evaluation apparatus, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0006] According to a first aspect of the embodiments of this specification, an object evaluation method is provided, comprising: Obtain the object parameters of the target object to be evaluated in terms of ontology attributes and interaction attributes; Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the corresponding feature analysis layer in the object evaluation model. The feature analysis layer is determined based on the attributes of the object parameters under the corresponding attribute dimension. The target ontology feature vector and the target interaction feature vector are fused through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object.

[0007] According to a second aspect of the embodiments of this application, an object evaluation apparatus is provided, comprising: The acquisition module is configured to acquire object parameters of the target object to be evaluated in the ontology attribute dimension and the interaction attribute dimension. The feature determination module is configured to obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension based on the object parameters under the ontology attribute dimension and the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model, wherein the feature analysis layer is determined based on the attributes of the object parameters under the corresponding attribute dimension. The object value determination module is configured to fuse the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object.

[0008] According to a third aspect of the embodiments of this application, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of the object evaluation method.

[0009] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided that stores computer-executable instructions that, when executed by a processor, implement the steps of the object evaluation method.

[0010] According to a fifth aspect of the present application, a chip is provided that stores a computer program, which, when executed by the chip, implements the steps of the object evaluation method.

[0011] According to a sixth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the object evaluation method described above.

[0012] The object evaluation method provided in this application comprehensively covers the inherent information and interactive association information of the target object by obtaining the object parameters of the target object under the ontology attribute dimension and the interaction attribute dimension respectively. Relying on the feature analysis layer in the object evaluation model that is adapted to each attribute dimension, it can extract accurate target ontology feature vectors and target interaction feature vectors based on the characteristics of object parameters under different attribute dimensions, effectively ensuring the adaptability and accuracy of feature extraction in each dimension. Then, through the fusion layer, the dual-dimensional feature vectors are fused to integrate the inherent ontology features and interactive association features of the target object, and finally obtain comprehensive, accurate and realistic target value evaluation information, realizing a comprehensive and accurate evaluation of the value of the target object. Attached Figure Description

[0013] Figure 1 This is a flowchart of an object evaluation method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a feature analysis layer provided in an embodiment of this application; Figure 3 This is a flowchart illustrating an object evaluation method applied in a game scene according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an object evaluation device provided in an embodiment of this application; Figure 5 This is a structural block diagram of a computing device provided in one embodiment of this application; Obviously, the accompanying drawings are only some illustrative examples of embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0015] The terminology used in one or more embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of one or more embodiments of this application. The singular forms “a,” “the,” and “the” used in one or more embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” used in one or more embodiments of this application refers to and includes any or all possible combinations of one or more associated listed items. For example, “A and / or B” can represent three cases: only A exists, only B exists, and both A and B exist, where A and B can be singular or plural. The character “ / ” generally indicates that the preceding and following objects are in an “or” relationship. “At least one of the following” or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of A, B, or C can represent: A, B, C, “A and B,” “A and C,” “B and C,” or “A and B and C,” where A, B, and C can be single or multiple.

[0016] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those described herein. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this application, and similarly, second may also be referred to as first. Furthermore, the terms "comprising," "having," "including," and variations thereof all indicate non-exclusive inclusion, such as a process, method, system, product, or device that comprises a series of steps or units, not limited to the steps or units explicitly listed, but may also include other steps or units not explicitly listed or inherent in the process itself.

[0017] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0018] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0019] Convolutional kernel: The core component in convolutional neural networks used to extract local features of data, often called a filter or feature detector. Essentially, it is a small, fixed-size matrix (e.g., 3×3, 5×5). During network operation, the convolutional kernel acts like a sliding window, moving row-by-row and column-by-column across the input data (e.g., images, time-series data), capturing local features through matrix operations. Different convolutional kernels can specifically extract different types of features such as edges, textures, corners, and contours, making it a key tool in deep learning for automatically discovering local patterns in data.

[0020] Fully connected layer neurons: These are the basic computational units of a fully connected layer in a neural network. Their most crucial characteristic is that each neuron is connected to all output nodes of the previous layer, hence the name "fully connected." Their main function is to integrate all local features extracted from preceding convolutional and pooling layers, fusing scattered local information into complete global features. Ultimately, they undertake the decision-making work for tasks such as classification and regression, and are the core unit for achieving the final output result of the neural network. Their disadvantage is a relatively large number of parameters.

[0021] Activation functions are key functions in neural networks that add non-linear capabilities to neurons. They perform a non-linear transformation on the weighted summation of neuron values ​​before outputting the result. Without activation functions, multi-layer neural networks are equivalent to simple linear models and cannot learn complex patterns and rules in data. By introducing non-linear characteristics, activation functions enable neural networks to fit arbitrarily complex data distributions, forming the foundation for the powerful learning capabilities of deep learning models. Common activation functions include ReLU, Sigmoid, Tanh, and Softmax.

[0022] It's important to note that Massively Multiplayer Online Role-Playing Games (MMORPGs) possess a distinct core characteristic of individual-group symbiosis. The construction and operation of their game ecosystem heavily rely on player communities, rather than simply emphasizing solo competition and individual combat. The core gameplay of this genre revolves around community-oriented content such as team-based dungeon raids, guild collaboration, factional social interactions, and factional warfare. Players establish game connections through frequent and continuous social interactions and collaborative behaviors, thereby forming a long-term sense of companionship and community belonging. This is the core key to maintaining player engagement and ensuring the game's lifespan.

[0023] However, MMORPGs have complex systems and long gameplay cycles, presenting new players with significant entry barriers in their initial experience. They struggle to quickly integrate into social circles like guilds, lack guidance from experienced players in core dungeon challenges, and often lack understanding of class skills, progression paths, and gameplay rules. This makes them highly susceptible to loneliness and high learning curve, leading to churn after a short period and severely hindering new user retention and ecosystem expansion. Statistics show that if new players receive guidance and support from experienced core players—specifically, core players with high community influence acting as mentors, actively guiding them through dungeons, explaining gameplay, and recruiting them into guild communities—the long-term retention rate can increase by more than three times compared to scenarios without guidance. Effective new player retention is a fundamental prerequisite for MMORPGs to achieve user expansion, maintain ecosystem vitality, and ensure long-term stable operation.

[0024] In terms of monetization, MMORPGs exhibit typical community-based monetization characteristics. Players' willingness to pay is highly dependent on social relationships and group behavior, with high-value core players becoming the trendsetters for monetization within the community. The spending behavior of these core players has a strong ripple effect: when they purchase new appearance items, mounts, costumes, etc., guild members and players within their social circles are likely to follow suit, engaging in group-wide monetization; when they participate in core paid content such as paid dungeons, season passes, and version packs, surrounding players will also proactively follow suit to avoid falling behind the team's progress and keep up with the community's gameplay pace. This community-based monetization model driven by core players, based on players' spontaneous willingness, has higher acceptance of payments and is less likely to trigger player resistance, making it a crucial path for game revenue conversion.

[0025] In terms of user activity and online rate, core influential players typically possess stronger gameplay comprehension and team leadership skills, enabling them to lead their teams to consistent victories in core gameplay modes such as dungeons and faction wars. This, in turn, earns them recognition and followers among other players within the community. The positive experience players gain from cooperative victories continuously motivates them to participate in the game again. Players will proactively agree on online team-up times, directly boosting key activity metrics such as daily average online users and monthly average active users, thus creating a positive cycle of community activity.

[0026] In conclusion, high-value core influencers are key to maintaining the MMORPG game community ecosystem, improving new user retention, and driving community spending and activity. Accurately identifying and managing such players has irreplaceable strategic value for the long-term operation of the game. However, existing player screening and identification methods are clearly one-sided, mostly focusing on superficial data such as character combat power, spending amount, and character level. Players selected through such methods often possess only individual gaming skills but lack community leadership and social influence, failing to effectively activate community vitality, boost overall activity among both new and old players, and truly leverage the positive driving force of core players on the game ecosystem.

[0027] To address the aforementioned technical problems, this specification provides an object evaluation method, as well as an object evaluation apparatus, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0028] See Figure 1 , Figure 1 A flowchart of an object evaluation method according to an embodiment of this specification is shown, specifically including the following steps 102-106.

[0029] Step 102: Obtain the object parameters of the target object to be evaluated in the ontology attribute dimension and the interaction attribute dimension.

[0030] The target object to be evaluated in the embodiments of this specification refers to the individual subject whose value and influence need to be determined in the operating ecosystem, and is the core processing object targeted by this evaluation process.

[0031] Among them, the ontology attribute dimension is used to describe the classification dimension to which various parameters inherent and independently exist in the target object belong. It only reflects the object's own basic state and inherent capabilities, and does not involve the related interaction behavior with other objects.

[0032] The interaction attribute dimension is used to describe the classification dimension of the parameters that the target object belongs to in relation to, linkage with, and transmission with other subjects, scenes or content within the ecosystem. It is used to reflect the object's relational performance and external influence capabilities within the ecosystem.

[0033] Object parameters refer to quantifiable raw data and behavioral indicators collected from the ontology attribute dimension and interaction attribute dimension. They are the basic data source for subsequent feature extraction and value assessment.

[0034] Object parameters under the ontology attribute dimension include, but are not limited to: their own basic values, cumulative participation time, continuous activity frequency, total amount of resources held, inherent attribute level, basic status indicators, and individual effectiveness level.

[0035] The object parameters under the interaction attribute dimension include, but are not limited to: the total number of objects affected, the percentage of the radiation range, the number of cross-groups reached, the increase in group interaction, the collaborative participation rate, the percentage of activity improvement, the number of objects reached by the information, the dissemination conversion rate, and the number of secondary diffusions.

[0036] In the context of gaming, the ontological attribute dimension is an assessment dimension that represents the inherent basic state, core capabilities, and stable participation of a game user. It serves as the basis for judging whether a user has the ability to influence a group and for filtering out users with no influence or low potential.

[0037] For example, in a game scenario, object parameters under the ontology attribute dimension may include, but are not limited to: character level parameters, character combat power parameters, average daily online time parameters, average monthly active days parameters, and total payment amount in the past 3 months parameters.

[0038] It should be noted that in game scenarios, directly collected raw game business data often has problems such as a large range of values, inconsistent numerical magnitudes, and different indicator units. If the raw data is directly used for subsequent feature extraction and value assessment, it is easy to cause an imbalance in feature weights and deviation in assessment results. Therefore, it is necessary to perform segmented quantization mapping and standardization processing on the collected raw data. The values ​​obtained after the above normalization processing are the object parameters under the ontology attribute dimension. Therefore, in the game scenario, the object parameters may include processed character level parameters, character combat power parameters, average daily online time parameters, average monthly active days parameters, and total spending in the past 3 months parameters. Among them, the character level parameter is obtained from the original character level data through segmented quantization mapping and standardization processing, and is used to reflect the user's game character's growth level; the character combat power parameter is obtained from the original character combat power data through corresponding mapping and standardization processing, and is used to characterize the user's game character's equipment and combat capabilities; the average daily online time parameter and the average monthly active days parameter are obtained from the original online time data and active days data, respectively, and are used to reflect the user's game participation frequency and stability; the total spending in the past 3 months parameter is obtained from the original spending data and is used to characterize the user's game spending level.

[0039] In the game scenario, the object parameters of the ontology attribute dimension mainly come from the user_basic table in the game database and related activity and payment logs.

[0040] For example, for the role level parameter, you can directly query the level field in the user_basic table to get the object's current level. Based on the corresponding mapping rules, the object's current level is mapped to obtain the mapped value.

[0041] For example, if the query finds an object with a level of 128, according to the rule "levels 111~129 = 0.8 (that is, levels between 111 and 129 are mapped to 0.8)," the resulting mapping value x_base is 0.8.

[0042] Substituting the mapping value 0.8 into the standardized formula "x=x_base*k+b", where k=0.5 and b=0 for the object parameter of the ontology attribute dimension, we get x=0.8*0.5+0, and thus the final feature value x=0.4.

[0043] k=0.5 means that the ontology attribute dimension has a low base weight in the model.

[0044] For example, if the query finds an object with a level of 108, according to the rule "if the user's level is lower than 110, the mapping value is 0", then the mapping value x_base is 0.

[0045] Optionally, objects with a character level parameter below a certain value can be marked as "objects with no potential" and excluded from subsequent analysis.

[0046] To determine a character's combat power (_Role_Power), you can directly query the `combat_power` field in the `user_basic` table. Compare the combat power value with a threshold. For example, if an object has a combat power of 225,000, according to the rule "210,000~220,000 = 0.8", the mapped value `x_base` is 0.8. Therefore, `x = 0.8 * 0.5 + 0 = 0.4`.

[0047] For the average daily online time (_Daily_Online_Time), the total online time (in seconds) for each day of the previous month can be calculated from the online time details log. Then, the average online time for all days is calculated to obtain the average daily online seconds, which is then converted to hours. The calculated average daily online hours (e.g., 5.5 hours) is compared with the rules. Based on "5~6 hours = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.5 + 0 = 0.4.

[0048] For the average monthly active days (_Monthly_Active_Days), count the number of days in the previous month when the object had a login record (after deduplication) from the object's login log. Compare the active days (e.g., 27 days) with the rule. Based on "26~28 days = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.5 + 0 = 0.4.

[0049] For the total payment amount over the past three months (_Three_Month_Pay), calculate the sum of all recharge amounts for this user within the past three calendar months from the recharge transaction record. Compare the total amount (e.g., 4500 yuan) with the rules. Based on "4000~5000 yuan = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.5 + 0 = 0.4.

[0050] In the context of gaming, the interaction attribute dimension is an evaluation dimension that characterizes the group interaction behaviors and external radiation capabilities of game users within the game ecosystem, including social interaction, gameplay-driven activities, word-of-mouth dissemination, and ecosystem traffic generation with other players, guilds, gameplay scenarios, and internal and external communities. This dimension focuses on users' non-inherent interactive behaviors and is the core basis for measuring the user group's radiation capability, ecosystem-driving value, social influence, and dissemination appeal. At the same time, the evaluation weight can be flexibly adjusted in combination with game operation activities and version update needs to adapt to the operational focus at different stages.

[0051] It should be noted that directly collected raw social, gameplay, and dissemination business data have issues such as diverse indicator types, inconsistent statistical standards, and significant differences in value weights. Furthermore, it needs to be adapted to the flexible adjustment requirements of game operation activities and version updates. Directly using raw data can easily lead to an imbalance in feature weights and fail to reflect actual operations. Therefore, it is necessary to perform segmented quantification mapping, weighted calculation, and standardization on the collected raw interaction data. The values ​​obtained after the above normalization process are the object parameters under the interaction attribute dimension.

[0052] For example, in a game scenario, the interaction attribute dimension can include social interaction attribute dimension, gameplay-driven attribute dimension, and word-of-mouth spread attribute dimension.

[0053] Among them, the object parameters under the social linkage attribute dimension may include, but are not limited to: guild position level, guild member activity correlation, frequency of guild activities, influence of sect channel speech, social circle size and cross-circle linkage capability.

[0054] The object parameters under the social interaction attribute dimension can be obtained from social_guild, chat_log, and social interaction logs.

[0055] For guild position levels (_Guild_Position), the social_guild table can be queried to obtain the user's position code within the current guild (e.g., 1 represents the guild leader). This position code is then directly mapped to a score. For example, if the position is "Guild Leader," the mapped value x_base is 1.0. x = 1.0 * 0.9 + 0 = 0.9. k = 0.9 reflects the high weight of social attributes in influence assessment.

[0056] For guild member activity correlation (_Guild_Active_Correlation), login logs and guild member relationships can be combined. First, identify all other members in the user's guild. Then, calculate the activity date list for the user and each other member in the previous month. Iterate through each other member and determine if the overlap of activity dates reaches a certain standard (e.g., more than half of the activity days are on the same day). Count the number of other members that meet this condition, divide by the total number of active guild members, and obtain the overlap percentage. For example, if the overlap percentage is 85%, according to the rule "80%-89%=0.8", the mapping value x_base is 0.8. x=0.8*0.9+0=0.72.

[0057] For the frequency of guild activities (_Guild_Activity_Frequency), we can count the number of times the user initiated activities as an organizer in the previous month from the guild activity organization log. Compare the number of events (e.g., 9 times) with the rule. Based on "9 - 10 times / month = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.9 + 0 = 0.72.

[0058] To assess the influence of a user's speech in the "Faction_Speech_Influence" channel, we can filter the `chat_log` table to find all their speech records in the "Faction" channel last month, and summarize the number of likes and replies to each speech. We then consider two conditions: 1. Total likes ≥ 80; 2. Number of unique users who replied ≥ 50. According to the rules, both conditions are met, resulting in a score of 1.0; one condition is met, resulting in 0.5; and neither condition is met, resulting in 0.2. For example, if a user has 120 total likes but only 40 replies, the mapping value `x_base` is 0.5. Therefore, `x = 0.5 * 0.9 + 0 = 0.45`.

[0059] For the Social Circle Scale, we can count the number of all other unique users who had effective interactions with the user in the previous month from various social interaction logs such as private chats, group chats, friend chats, and transactions. We then compare the number of interactions (e.g., 65 people) with the rules. Based on "60-70 people = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.9 + 0 = 0.72.

[0060] For cross-circle linkage capability (_Cross_Circle_Linkage), "circle" is defined as guilds, fixed dungeon teams, battlefield squads, interest groups, etc. Two things can be counted from the user's activity logs across various circles: 1. The number of different circle types the user was active in last month; 2. The number of times the user interacted across different circles (e.g., calling fixed team members in the guild channel for a dungeon run). Two conditions are combined for judgment: 1. Number of active circles ≥ 3; 2. Number of cross-circle linkage actions ≥ 15. Both conditions are met, resulting in a value of 1.0; one condition is met, resulting in 0.5; neither condition is met, resulting in 0.2. For example, if a user is active in guilds, fixed teams, and battlefield squads, but only has 10 cross-circle linkage actions, then the mapping value x_base is 0.5. x = 0.5 * 0.9 + 0 = 0.45.

[0061] Among them, the object parameters under the gameplay-driven attribute dimension may include, but are not limited to: frequency of leading teams in core dungeons, completion rate of core dungeons, appeal of faction wars, number of new players guided, and frequency of gameplay tutorials for new players.

[0062] The object parameters under the attribute dimension driven by the gameplay can be obtained from the behavior_instance table and the mentor-apprentice and teaching logs.

[0063] For the frequency of leading core dungeons (_Dungeon_Lead_Frequency), we can use the behavior_instance table to count the number of core dungeons the user successfully completed in the previous month as the "leader". Compare the number of leading dungeons (e.g., 12 times) with the rules. Based on "11-15 times / month = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.9 + 0 = 0.72.

[0064] For the Core Dungeon Clear Rate (_Dungeon_Clear_Rate), you can find the total number of times a user participated in Core Dungeons and the number of times they successfully cleared them in the previous month from the behavior_instance table. The clear rate is calculated as: Clear Rate = Number of Successful Dungeons / Total Number of Dungeons. For example, if the clear rate is 92%, according to the rule "≥90%=1.0", the mapped value x_base is 1.0. x = 1.0 * 0.9 + 0 = 0.9.

[0065] For Campaign Call Power (_Camp_Call_Power), we can calculate it from the Campaign rally or registration logs by counting the number of times the user initiated rallies as a rally leader in the previous month, and the average number of participants successfully mobilized per rally. We combine two conditions: 1. Number of rallies ≥ 8; 2. Average number of participants mobilized ≥ 50. Meeting both conditions earns 1.0, meeting one earns 0.5, and not meeting either earns 0.2. For example, if a user initiates rallies 10 times a month, but only mobilizes an average of 30 people per rally, then the mapping value x_base is 0.5. x = 0.5 * 0.9 + 0 = 0.45.

[0066] For the number of new player referrals (_Newbie_Guide_Number), the number of different new players successfully referred by the user in the previous month can be counted from the mentor-apprentice system or the new player referral behavior log. Compare the number of referred players (e.g., 13) with the rules. Based on "11-15 people / month = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.9 + 0 = 0.72.

[0067] To determine the frequency of tutorials for new players (_Newbie_Teach_Frequency), we can calculate the number of times a user performed tutorials in the previous month from dedicated tutorial logs (such as explanations of dungeon mechanics and gear setups). Compare this number of tutorials (e.g., 18 times) with the established rules. Based on "16-20 times / month = 0.8", the mapping value x_base is 0.8. Therefore, x = 0.8 * 0.9 + 0 = 0.72.

[0068] Among them, the object parameters under the word-of-mouth dissemination attribute dimension may include, but are not limited to: the frequency of positive comments in the game, the proportion of positive reviews from surrounding users, the effect of positive content output outside the game, and the effect of external traffic generation.

[0069] The object parameters under the word-of-mouth dissemination attribute dimension can be obtained from in-game chat logs, player feedback systems, external community data, and traffic acquisition records.

[0070] To measure the frequency of positive in-game comments (_InGame_Positive_Speech), we can filter the user's comments in public channels such as World, Guild, and Sect from the chat_log table last month, and identify the comments marked as "positive" using keywords or a simple sentiment model. We then count the number of positive comments (e.g., 45). Based on "41-50 times / month = 0.8", we obtain a mapping value x_base of 0.8. x = 0.8 * 0.7 + 0 = 0.56. k = 0.7 indicates that the reputation attribute weight is between personal base and social / gameplay attributes.

[0071] To determine the percentage of positive reviews from surrounding users (_Surrounding_Positive_Rate), we can obtain all user reviews of that user from the player's review, like, and report feedback system, and identify their sentiment (positive / negative / neutral). The percentage of positive reviews is calculated as: Number of positive reviews / Total number of reviews. For example, if 20 reviews are received, and 18 are positive (90%), according to the rule "≥90%=1.0", the mapped value x_base is 1.0. Therefore, x = 1.0 * 0.7 + 0 = 0.7.

[0072] For the positive content output effect outside the game (_OutGame_Positive_Content), the number of game-related posts, videos, and other content published by the user account, as well as interaction data (likes, shares), can be obtained from the associated external community data. Two conditions are considered: 1. Number of posts ≥ 5; 2. Total interactions (likes + shares) ≥ 200. Meeting both conditions yields a value of 1.0, meeting one yields 0.5, and not meeting either yields 0.2. For example, if a user publishes 3 high-quality videos with a total interaction count of 500, the mapping value x_base is 0.5. x = 0.5 * 0.7 + 0 = 0.35.

[0073] For the external traffic generation effect (_External_Drainage_Effect), count the number of new users who successfully registered and created a role in the previous month through the user's exclusive link or invitation code from user invitations or traffic generation. Compare the number of traffic generated (e.g., 9 people) with the rules. Based on "9-10 people / month = 0.8", the mapping value x_base is 0.8. x = 0.8 * 0.7 + 0 = 0.56.

[0074] Step 104: Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model. The feature analysis layer is determined based on the attributes of the object parameters under the corresponding attribute dimension.

[0075] In an optional implementation of this embodiment, the interaction attribute dimension includes an object radiation sub-dimension, an object driving sub-dimension, and a positive information diffusion sub-dimension. The object radiation sub-dimension characterizes the range of other objects that the target object can influence, the object driving sub-dimension characterizes the effect of the target object on improving the activity of other objects, and the positive information diffusion sub-dimension characterizes the effect of the target object on spreading positive information. Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the corresponding feature analysis layer in the object evaluation model, including: Based on the object parameters under the ontology attribute dimension, the target ontology feature vector under the ontology attribute dimension is obtained through the feature analysis layer corresponding to the ontology attribute dimension. Based on the object parameters under the object radiation sub-dimension, object driving sub-dimension, and positive information diffusion sub-dimension, the target interaction feature vector under the interaction attribute dimension is obtained through the feature analysis layer corresponding to the interaction attribute dimension. In the object evaluation model, one feature analysis layer is used to process the object parameters of at least one sub-dimension under the interaction attribute dimension.

[0076] Among them, the object radiation sub-dimension is a sub-dimension under the interaction attribute dimension, used to characterize the social interaction behavior characteristics of the target object within the ecosystem, such as social association, circle linkage, and guild interaction.

[0077] The object-driven sub-dimension is a sub-dimension under the interaction attribute dimension used to represent the collective influence of a target object on other objects in aspects such as gameplay organization, new player guidance, and faction appeal.

[0078] The positive information diffusion sub-dimension is a sub-dimension under the interaction attribute dimension, used to characterize the positive information transmission behavior of the target object, such as positive word-of-mouth dissemination and external traffic generation, both inside and outside the ecosystem.

[0079] Specifically, the object radiation sub-dimension is used to characterize the range of other objects that the target object can influence, to depict the breadth of the target object's influence in its system, and to reflect the number of objects that the target object can reach, the distribution boundary, and the extent of the influence. It is an important dimension for measuring the scope of influence and the basis of its transmission.

[0080] The object parameters under the object radiation sub-dimension may include, but are not limited to: the total number of objects that the target object has come into contact with, the object distribution density within the scope of influence, the proportion of effectively covered objects in the overall system, the number of radiation coverages across regions / groups, and the number of objects with stable influence.

[0081] The object-driven sub-dimension is used to characterize the effect of the target object on improving the activity of other objects. It is used to depict the positive promoting effect of the target object on the participation, interaction frequency and operation status of other objects in the system, and to reflect the actual performance of the target object in driving the overall system to remain active and improve the degree of collaboration.

[0082] Object parameters under the object-driven sub-dimension may include, but are not limited to: the increase in the interaction frequency of other objects, the incremental proportion of collaborative participation behavior, the overall activation rate of objects within the system, the proportion of objects passively responding to the guidance of the target object, the increase in running intensity and activity frequency, and the improvement level of the completion of group collaborative actions.

[0083] The positive information diffusion sub-dimension is used to characterize the effect of the target object on the dissemination of positive information. It is used to depict the effectiveness of the target object in transmitting, promoting and reaching more objects with positive information, and to reflect the target object's ability to expand the coverage of positive information and enhance the efficiency of information transmission.

[0084] The object parameters under the positive information diffusion sub-dimension may include, but are not limited to: the number of objects reached by positive information, the rate of expansion of the coverage of information dissemination, the proportion of objects that receive information and generate a response, the transmission depth of cross-node information transmission, the number of secondary diffusions of positive information, and the effective reach rate of information dissemination.

[0085] Using the previous example, in the game scenario, the object radiation sub-dimension in the interaction attribute dimension corresponds to the social linkage attribute, which is used to characterize the characteristics of game users' social interaction behaviors such as guild interaction, social circle linkage, and channel speech.

[0086] The object-driven sub-dimension is a sub-dimension under the interaction attribute dimension, corresponding to gameplay-driven attributes, and is used to characterize the characteristics of game users' group gameplay-driven behaviors such as leading dungeons, new player guidance, and faction rallying.

[0087] The positive information diffusion sub-dimension is a sub-dimension under the interaction attribute dimension, corresponding to the word-of-mouth propagation attribute. It is used to characterize the positive information diffusion behavior characteristics of game users inside and outside the ecosystem, such as positive statements, word-of-mouth evaluations, and external traffic.

[0088] The feature analysis layer is a functional processing layer in the object evaluation model that is specifically responsible for feature extraction, feature transformation, and vector generation of object parameters in the corresponding dimension or sub-dimension. The target ontology feature vector is vector data that condenses and represents the inherent basic attribute features of the target object itself after being processed by the feature analysis layer dedicated to the ontology attribute dimension. The target interaction feature vector is vector data that integrates three types of sub-dimension features: object radiation, object driving, and positive information diffusion after being processed by the feature analysis layer corresponding to the interaction attribute dimension.

[0089] This solution first refines the interaction attribute dimension, dividing it into three sub-dimensions: object radiation, object-driven, and positive information diffusion. In the feature extraction stage, a dimensional adaptation processing logic is adopted. First, for the object parameters under the ontology attribute dimension, the corresponding feature analysis layer for that dimension is called for processing, thereby obtaining a target ontology feature vector that accurately reflects the inherent characteristics of the target object. Then, for the object parameters of the three sub-dimensions under the interaction attribute dimension, the corresponding feature analysis layer for that dimension is used for processing. A single feature analysis layer in the object evaluation model can handle the processing task of at least one sub-dimension object parameter under the interaction attribute dimension. Finally, the results are integrated to generate a target interaction feature vector that comprehensively reflects the interactive behavior characteristics of the target object, completing the independent extraction and generation of dual-dimensional feature vectors.

[0090] Using the previous example, in the game scenario, the object parameters under the ontology attribute dimension are parameters such as character level, character combat power, and average daily online time after quantization mapping and standardization. Through the feature analysis layer corresponding to the ontology attribute dimension, feature extraction and vector transformation are performed on these parameters to obtain the target ontology feature vector that represents the basic ability and participation of the game user.

[0091] Within the interaction attribute dimension, the object radiation sub-dimension corresponds to the social linkage attribute dimension, which includes object parameters such as guild rank and social circle size. The object driving sub-dimension corresponds to the gameplay-driven attribute dimension, which includes object parameters such as core dungeon team frequency and new player guidance number. The positive information diffusion sub-dimension corresponds to the word-of-mouth dissemination attribute dimension, which includes object parameters such as in-game positive speech frequency and external traffic generation effect. Through the feature analysis layer corresponding to the interaction attribute dimension in the object evaluation model, the object parameters of the above three attribute dimensions can be processed individually or in combination, ultimately generating a target interaction feature vector that integrates the game user's social linkage ability, group gameplay-driven ability, and positive word-of-mouth dissemination ability.

[0092] In the embodiments of this specification, by refining the interaction attribute dimension into three sub-dimensions—the object radiation sub-dimension, the object-driven sub-dimension, and the positive information diffusion sub-dimension—it is possible to comprehensively cover all kinds of core features of the object, avoiding the problem of one-sided representation of interaction features. By adopting a processing method of matching the corresponding feature analysis layer with the dimension, each feature analysis layer can be specifically adapted to the feature rules of the corresponding attribute dimension parameters, effectively improving the accuracy and adaptability of feature extraction. At the same time, a single feature analysis layer can handle the setting of attribute dimension parameters corresponding to at least one sub-dimension, which can flexibly adjust the model processing structure. While ensuring the quality of feature extraction, it optimizes the model running efficiency, providing comprehensive and reliable feature data support for subsequent fusion of dual-dimensional features and generation of accurate user value assessment information that fits the game scenario.

[0093] In one optional implementation of this embodiment, a feature analysis layer in the object evaluation model is used to process object parameters in a sub-dimension under the interaction attribute dimension; the target interaction feature vector includes the object radiation feature vector under the object radiation sub-dimension, the object driving feature vector under the object driving sub-dimension, and the positive information feature vector under the positive information diffusion sub-dimension. Based on object parameters under the object radiation sub-dimension, object-driven sub-dimension, and positive information diffusion sub-dimension, the target interaction feature vector under the interaction attribute dimension is obtained through the feature analysis layer corresponding to the interaction attribute dimension, including: Based on the object parameters under the object radiation sub-dimension, the object radiation feature vector under the object radiation sub-dimension is obtained through the feature analysis layer corresponding to the object radiation sub-dimension. Based on the object parameters under the object-driven sub-dimension, the object-driven feature vector under the object-driven sub-dimension is obtained through the feature analysis layer corresponding to the object-driven sub-dimension. Based on the object parameters under the positive information diffusion sub-dimension, the positive information feature vector under the positive information diffusion sub-dimension is obtained through the feature analysis layer corresponding to the positive information diffusion sub-dimension.

[0094] Among them, the object radiation feature vector refers to the feature vector obtained by extracting features of the object parameters under the sub-dimension through the feature analysis layer corresponding to the object radiation sub-dimension.

[0095] Object-driven feature vectors refer to feature vectors obtained by extracting features from object parameters in a sub-dimension through the feature analysis layer corresponding to that sub-dimension.

[0096] The positive information feature vector refers to the feature vector obtained by extracting features from the object parameters under the positive information diffusion sub-dimension through the feature analysis layer corresponding to the sub-dimension.

[0097] This scheme explicitly defines the feature extraction structure of the object evaluation model. In the object evaluation model, a feature analysis layer is only used to process the object parameters of one sub-dimension under the interaction attribute dimension. The target interaction feature vector is composed of the object radiation feature vector, the object driving feature vector, and the positive information feature vector.

[0098] During feature extraction, the four feature extraction branches can be executed in parallel, specifically: Based on object parameters under the ontology attribute dimension, the target ontology feature vector is obtained through the feature analysis layer corresponding to the ontology attribute dimension. At the same time, for the three sub-dimensions under the interaction attribute dimension, based on object parameters under the object radiation sub-dimension, the object radiation feature vector is obtained through the feature analysis layer corresponding to the sub-dimension; based on object parameters under the object driving sub-dimension, the object driving feature vector is obtained through the feature analysis layer corresponding to the sub-dimension; and based on object parameters under the positive information diffusion sub-dimension, the positive information feature vector is obtained through the feature analysis layer corresponding to the sub-dimension.

[0099] After the four branches complete feature extraction in parallel, the feature vectors of the three sub-dimensions are combined to form a complete target interaction feature vector. This parallel processing method can effectively improve the efficiency and accuracy of feature extraction, avoiding problems such as excessive time consumption and feature interference caused by serial processing.

[0100] In the game scenario, the entity attribute dimension includes object parameters such as character level, character combat power, and average daily online time after quantitative mapping and standardization. The object radiation sub-dimension (corresponding to the social linkage attribute dimension) includes object parameters such as guild position level and social circle size. The object driving sub-dimension (corresponding to the gameplay-driven attribute dimension) includes object parameters such as the frequency of leading core dungeons and the number of new players guided. The positive information diffusion sub-dimension (corresponding to the word-of-mouth spread attribute dimension) includes object parameters such as the frequency of positive statements in the game and the effect of external traffic generation.

[0101] During feature extraction, the four branches mentioned above can be executed in parallel: the feature analysis layer corresponding to the ontology attribute dimension processes its own object parameters synchronously, and the feature analysis layers corresponding to each of the three sub-dimensions process the object parameters of their respective sub-dimensions. There is no need to wait for one branch to complete before executing the next branch. The target ontology feature vector, object radiation feature vector, object driving feature vector, and positive information feature vector are obtained synchronously. Finally, the feature vectors of the three interaction sub-dimensions are integrated to obtain the target interaction feature vector of the game user, which greatly shortens the feature extraction time and avoids mutual interference between feature extraction of different branches.

[0102] In the embodiments of this specification, an independent processing method is adopted, with one sub-dimension corresponding to one feature analysis layer. Combined with the execution logic of four feature extraction branches, on the one hand, each feature analysis layer can be fully adapted to the parameter characteristics of the corresponding sub-dimension and its attribute dimension, effectively avoiding mutual interference between different types of interactive features during the extraction process, and greatly improving the accuracy and targeting of feature extraction; on the other hand, the four feature extraction branches can be processed in parallel, so that feature extraction of the ontology attribute dimension and the three interactive sub-dimensions can be promoted simultaneously, significantly shortening the overall time consumption of feature extraction, improving the model processing efficiency, and solving the problems of long processing time and low efficiency of serial processing.

[0103] In one optional implementation of this embodiment, the feature analysis layer includes a feature extraction module, a feature fusion module, and an output module; Based on object parameters under the ontology attribute dimension and interaction attribute dimension, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the corresponding feature analysis layer in the object evaluation model, including: Based on object parameters under the ontology attribute dimension and interaction attribute dimension, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are extracted through the feature extraction module in the corresponding feature analysis layer. Based on at least one target local ontology feature and at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the feature fusion module in the corresponding feature analysis layer. Based on the fusion of ontology features and fusion of interaction features, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the output module in the corresponding feature analysis layer.

[0104] The feature extraction module is the foundational functional module of the feature analysis layer. Its core function is to filter and extract preliminary local features that reflect the core characteristics of the parameters from the original input parameters, providing a basis for subsequent feature integration. The implementation method can be flexibly selected, and there are two examples: Example 1 uses a combination of convolutional layers and activation functions to capture the local correlation features of the parameters through convolution operations, and the activation function implements nonlinear transformation to enhance the feature expression capability; Example 2 uses a combination of recurrent neural networks (RNN) and activation functions to extract the sequence features of the parameters through temporal modeling, adapting to input parameters with temporal characteristics.

[0105] The feature fusion module is an intermediate functional module of the feature analysis layer. Its core function is to integrate, normalize, and optimize multiple local features output by the feature extraction module, eliminating the one-sidedness of individual local features and forming a fused feature that can comprehensively reflect the overall characteristics of the input parameters. There are two examples of its implementation: Example 1 uses a combination of fully connected layers and activation functions. The fully connected operation maps multiple local features to the same feature space, and the activation function further enhances the non-linear expression of the fused feature. Example 2 uses a combination of attention mechanism and fully connected layers. The attention mechanism assigns different weights to different local features to highlight the core local features, and then the final fusion is completed through the fully connected layer.

[0106] The output module is the final functional module of the feature analysis layer. Its core function is to perform format conversion and standardization on the fused features output by the feature fusion module, and finally output feature vectors that meet the requirements of subsequent model processing. There are two implementation examples: Example 1 uses a fully connected layer in the output layer to directly map the fused features into feature vectors of fixed dimensions, ensuring that the vector format is uniform; Example 2 uses a combination of a fully connected layer and a normalization layer. After completing the vector mapping, the normalization process eliminates the difference in magnitude of the feature vectors and improves the stability of subsequent processing.

[0107] Local features are preliminary features extracted from the original parameters by the feature extraction module. A single local feature corresponds to a core characteristic of the original parameters, and multiple local features together constitute the preliminary feature set of the original parameters.

[0108] Fusion features are comprehensive features obtained by integrating multiple local features by the feature fusion module. They can comprehensively and uniformly represent the overall features of the input original parameters and are the basis for generating the final feature vector.

[0109] The feature vector is the final result output by the output module. It is a standardized vector expression of the original parameters after feature extraction and fusion. It can be directly used for subsequent processing of the model (such as feature fusion, value evaluation, etc.).

[0110] In this scheme, firstly, the original parameters to be processed are input into the feature analysis layer, where the feature extraction module performs preliminary processing on the original parameters. Based on the characteristics of the original parameters, an appropriate implementation method (such as convolutional layer + activation function, RNN + activation function, etc.) is selected to extract multiple local features that can reflect the core characteristics of the original parameters. Subsequently, all the extracted local features are input into the feature fusion module. This module can also select various implementation methods (such as fully connected layer 1 / 2 + activation function, attention mechanism + fully connected layer, etc.) to integrate and normalize multiple local features, eliminate the limitations of individual local features, and generate fused features that can comprehensively represent the overall characteristics of the original parameters. Finally, the fused features are input into the output module. Through the processing of the output module (such as output layer fully connected layer, fully connected layer + normalization layer, etc.), the format conversion and standardization of the feature vector are completed, and finally, a feature vector that meets the requirements of subsequent model processing is output. Throughout the process, the implementation methods of each module can be flexibly selected without being fixed or uniform. The core is to ensure the accuracy of feature extraction, the comprehensiveness of feature fusion, and the standardization of feature vectors through the division of labor and cooperation of the three-level modules, so as to provide high-quality feature support for subsequent model processing.

[0111] Example 1: Feature extraction module: It adopts the "convolutional layer + activation function" approach. It performs convolution operation on the original parameters to be processed, captures the local correlation features of the parameters, and then uses an activation function (such as ReLU) to achieve non-linear transformation, filter out invalid features, and finally extracts 3 local features (corresponding to the 3 core features of the original parameters respectively). Feature fusion module: The implementation method of "fully connected layer 1 + activation function" is adopted. The above three local features are input into fully connected layer 1, and the local features are mapped to the same feature space through fully connected operation. Then, the fusion effect is enhanced by activation function to generate a comprehensive fused feature. Output module: It adopts the implementation method of "output layer fully connected layer", which maps the fused features into the output layer fully connected layer as a feature vector of fixed dimension (such as 16 dimensions) to complete the final output.

[0112] Example 2: Feature extraction module: It adopts the implementation method of "recurrent neural network (RNN) + activation function" to perform time series modeling on the original parameters of the time series class to be processed, extract the sequence association features of the parameters, optimize the feature expression through activation function, and finally extract 4 local features; Feature fusion module: It adopts the implementation method of "attention mechanism + fully connected layer 2". The attention mechanism assigns weights to four local features (the core local features are given high weights), and then the fully connected layer 2 completes the feature integration to generate one fused feature. Output module: It adopts the implementation method of "fully connected layer + normalization layer". The fused features are input into the fully connected layer to complete the vector mapping, and then the normalization layer eliminates the difference in vector magnitude and outputs the standardized feature vector.

[0113] The two examples have different module implementation methods, but both complete the transformation from original parameters to feature vectors through the collaborative work of three modules: feature extraction, feature fusion, and output, thus verifying the flexibility and feasibility of the solution.

[0114] In the embodiments of this specification, by dividing the feature analysis layer into a feature extraction module, a feature fusion module, and an output module, a clear division of labor in the feature processing flow is achieved. This enables the orderly completion of local feature mining, multi-feature integration, and standardized output of feature vectors, effectively ensuring the accuracy and standardization of feature processing. Simultaneously, each module supports multiple flexible implementation methods. For example, the feature extraction module can use convolutional layers with activation functions or recurrent neural networks with activation functions; the feature fusion module can use fully connected layers with activation functions or attention mechanisms combined with fully connected layers; and the output module can use fully connected output layers or fully connected layers combined with normalization layers. These can be adapted to the characteristics of different input parameters, significantly improving the scenario adaptability of the feature analysis layer. The modular structure also allows each module to be independently optimized and replaced, reducing model design and optimization costs and enhancing model scalability. The final output feature vectors can accurately and comprehensively represent the core features of the input parameters, providing stable and high-quality feature data support for subsequent processing stages of the object evaluation model.

[0115] In one optional implementation of this embodiment, the feature extraction module includes a feature extraction unit and a pre-nonlinear mapping unit; Based on object parameters under the ontology attribute dimension and interaction attribute dimension, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are extracted through the feature extraction module in the corresponding feature analysis layer, including: Based on object parameters under the ontology attribute dimension and the interaction attribute dimension, at least one initial local ontology feature under the ontology attribute dimension and at least one initial local interaction feature under the interaction attribute dimension are extracted through the corresponding feature extraction unit. Based on at least one initial local ontology feature and at least one initial local interaction feature, at least one target local ontology feature in the ontology attribute dimension and at least one target local interaction feature in the interaction attribute dimension are obtained through the corresponding pre-nonlinear mapping unit.

[0116] The feature extraction unit is the core subunit of the feature extraction module. Its core function is to directly perform preliminary feature filtering and extraction on the object parameters under the input ontology attribute dimension and interaction attribute dimension to obtain unoptimized initial local features, which is the foundation of local feature extraction. Its implementation method can be flexibly selected, and there are two examples: Example 1, using a convolutional layer, captures the local correlation features of the parameters through convolution operations to quickly extract the core initial features; Example 2, using a feature selector (such as the ReliefF algorithm) to filter out the initial features that are highly relevant to the evaluation target from the input parameters and remove redundant information.

[0117] The pre-nonlinear mapping unit is an auxiliary sub-unit of the feature extraction module, connecting to the feature extraction unit. Its core function is to perform nonlinear transformation and optimization on the initial local features output by the feature extraction unit, eliminating linear redundancy of the initial features, enhancing the nonlinear expressive power of the features, and converting the initial local features into target local features that better meet the needs of subsequent model processing. There are two implementation examples: Example 1, using a single activation function (such as ReLU, Sigmoid) to perform nonlinear mapping on the initial local features, filtering out invalid features and enhancing feature discriminativeness; Example 2, using a combination of activation function and nonlinear transformation layer, first achieving preliminary nonlinear transformation through activation function, and then further optimizing feature distribution through nonlinear transformation layer to improve feature quality.

[0118] Initial local ontology features are preliminary local features extracted by the feature extraction unit from the object parameters of the ontology attribute dimension without nonlinear optimization. They only reflect the basic core characteristics of the ontology attribute parameters and have certain linear redundancy or insufficient expression problems.

[0119] The initial local interaction features are preliminary local features extracted by the feature extraction unit from the object parameters of the interaction attribute dimension without nonlinear optimization. They only reflect the basic core characteristics of the interaction attribute parameters and need to be further optimized before they can be used as the basis for subsequent fusion.

[0120] The core of this scheme is to further refine and decompose the feature extraction module, clearly defining it as consisting of a feature extraction unit and a pre-nonlinear mapping unit. Through the collaborative work of these two sub-units, accurate extraction and optimization from object parameters to target local features are achieved. The specific process is as follows: First, the object parameters under the ontology attribute dimension and the interaction attribute dimension are input into the feature extraction module of the corresponding feature analysis layer, respectively. Then, through the feature extraction unit within the feature extraction module, preliminary feature extraction is performed on the object parameters of the two dimensions, obtaining at least one initial local ontology feature under the ontology attribute dimension and at least one initial local interaction feature under the interaction attribute dimension. This process mainly achieves the preliminary screening of core features, retaining the basic features related to evaluation. Finally, the extracted initial local ontology features and initial local interaction features are input into the corresponding pre-nonlinear mapping unit, respectively. Through the nonlinear transformation and optimization processing of this unit, the linear redundancy of the initial local features is eliminated, and the nonlinear expression of the features is strengthened. Finally, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are obtained, providing a high-quality, high-discrimination local feature foundation for the subsequent integration work of the feature fusion module. Throughout the process, the feature extraction unit and the pre-nonlinear mapping unit have a clear division of labor and work together to ensure the comprehensiveness of the initial feature extraction and to optimize and improve the feature quality.

[0121] For example, to clearly demonstrate the collaborative workflow and implementation flexibility of the two sub-units of the feature extraction module, two different unit implementation combinations are given in conjunction with object parameter input scenarios, as follows: Example 1: Input parameters: Object parameters under the ontology attribute dimension (such as quantified basic indicators) and object parameters under the interaction attribute dimension (such as quantified interaction behavior indicators). Feature extraction unit: Implemented using convolutional layers, it performs convolution operations on object parameters of two dimensions respectively, captures the local correlation features of the parameters, and extracts 2 initial local ontology features (corresponding to the two core features of the ontology parameters) and 3 initial local interaction features (corresponding to the three core features of the interaction parameters). Pre-nonlinear mapping unit: Implemented using the ReLU activation function, it performs nonlinear mapping on the above 5 initial local features, filters out invalid features with negative values ​​and strengthens the expression of valid features, and finally obtains 2 target local ontology features and 3 target local interaction features, thus completing the processing of the feature extraction module.

[0122] Example 2: Input parameters: object parameters under the ontology attribute dimension and object parameters under the interaction attribute dimension; Feature extraction unit: Implemented using ReliefF feature selector, it selects features with high relevance to the subsequent evaluation target from the two-dimensional object parameters, extracts 3 initial local ontology features and 2 initial local interaction features, and removes redundant and irrelevant features; The pre-nonlinear mapping unit is implemented by combining the Sigmoid activation function and the nonlinear transformation layer. First, the initial local features are mapped to the 0-1 interval by the Sigmoid activation function, and then the feature distribution is adjusted by the nonlinear transformation layer to optimize the discriminative power of the features. Finally, three target local ontology features and two target local interaction features are obtained, completing the processing of the feature extraction module.

[0123] The two examples have different unit implementation methods, but both complete the transformation from object parameters to target local features through the initial extraction of feature extraction units and the optimization of pre-nonlinear mapping units, verifying the flexibility and feasibility of the scheme.

[0124] In the embodiments of this specification, by splitting the feature extraction module into a feature extraction unit and a pre-nonlinear mapping unit, a hierarchical processing of local features from "preliminary extraction to optimization and improvement" is realized. The feature extraction unit can specifically capture the core initial features of the input parameters, ensuring the comprehensiveness and relevance of the initial features and avoiding the omission of core features. The pre-nonlinear mapping unit effectively eliminates the linear redundancy of the initial local features and enhances the nonlinear expressive ability of the features through nonlinear transformation, solving the problems of insufficient initial feature expression and low discriminative power, and greatly improving the quality and adaptability of the target local features.

[0125] In one optional implementation of this embodiment, the feature fusion module includes a primary feature fusion submodule and a high-level feature fusion submodule; Based on at least one target local ontology feature and at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the feature fusion module in the corresponding feature analysis layer, including: Based on at least one target local ontology feature and at least one target local interaction feature, the primary fused ontology feature under the ontology attribute dimension and the primary fused interaction feature under the interaction attribute dimension are obtained through the corresponding primary feature fusion submodule. Based on at least one target local ontology feature and at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the corresponding high-level feature fusion submodule.

[0126] Primary Feature Fusion Submodule: The foundational submodule of the feature fusion module. Its core function is to perform preliminary integration of the target local features (target local ontology features and target local interaction features) output by the feature extraction module, completing the basic-level feature fusion and outputting preliminary fused features to provide a foundation for advanced feature fusion. Implementation examples: Example 1: Using feature concatenation + simple fully connected layer to quickly integrate multiple sets of local features; Example 2: Using average pooling fusion to perform preliminary normalization and integration of local features.

[0127] Advanced Feature Fusion Submodule: The core submodule of the feature fusion module, connecting to the primary feature fusion submodule (or directly connecting to the feature extraction module). Its core function is to perform more accurate and in-depth fusion of target local features, optimize the fusion effect by combining the correlation and importance of features, and output the final usable fused features. Implementation examples: Example 1, using an attention mechanism + fully connected layer to highlight the weight of core local features; Example 2, using residual fusion + nonlinear transformation to enhance the expressive power of the fused features.

[0128] Primary fusion ontology features: These are fused features obtained by the primary feature fusion submodule after performing preliminary fusion of the target local ontology features. They only complete basic integration and have not undergone in-depth optimization, but are basic comprehensive features with ontology attributes.

[0129] Primary fusion interaction features: These are fused features obtained by the primary feature fusion submodule after preliminary fusion of the target local interaction features. They complete basic integration, have not undergone in-depth optimization, and possess basic comprehensive features with interaction attributes.

[0130] Target Fusion Ontology Features: The final fused features are obtained by the advanced feature fusion submodule after deep fusion of the target local ontology features. After precise optimization, they can comprehensively and accurately represent the overall features of the ontology attributes and are used for vector transformation in the subsequent output module.

[0131] Target fusion interaction features: The final fusion features are obtained by the advanced feature fusion submodule after deep fusion of the target local interaction features. After precise optimization, they can comprehensively and accurately represent the overall features of the interaction attributes and are used for vector transformation in the subsequent output module.

[0132] This solution achieves hierarchical feature fusion of "basic fusion - deep optimization" through the collaborative cooperation of two sub-modules. The specific process is as follows: First, the target local ontology features (at least one) and target local interaction features (at least one) output by the feature extraction module are input into the corresponding feature fusion modules of the feature analysis layer. Subsequently, two parallel paths are processed simultaneously: On the one hand, the primary feature fusion sub-module performs preliminary integration of the two types of target local features, completing basic feature splicing and normalization to obtain primary fused ontology features and primary fused interaction features, achieving preliminary summarization of local features; on the other hand, the advanced feature fusion sub-module performs deep fusion of the two types of target local features, combining feature correlation and importance for weight allocation and redundancy elimination, completing deep feature optimization, and finally obtaining target fused ontology features under the ontology attribute dimension and target fused interaction features under the interaction attribute dimension. Throughout the process, the primary sub-module is responsible for "laying the foundation" and completing basic fusion; the advanced sub-module is responsible for "quality improvement" and completing deep optimization. The two work together to improve the comprehensiveness and accuracy of the fused features, providing a high-quality fused feature foundation for the subsequent output module to generate feature vectors.

[0133] To clearly demonstrate the collaborative workflow and implementation flexibility of the primary and advanced submodules, two different examples of submodule implementation combinations are given below: Example 1: Input features: 2 target local ontology features (such as core local features corresponding to ontology parameters) and 3 target local interaction features (such as core local features corresponding to interaction parameters). The primary feature fusion submodule is implemented using a "feature concatenation + simple fully connected layer". It concatenates the local ontology features of two targets and integrates them through a simple fully connected layer to obtain a primary fused ontology feature; it also concatenates the local interaction features of three targets and integrates them through a simple fully connected layer to obtain a primary fused interaction feature. Advanced Feature Fusion Submodule: Implemented using "attention mechanism + fully connected layer", it assigns weights to the local ontology features of two targets (with core features given high weights), and then deeply integrates them through a fully connected layer to obtain a fused ontology feature of one target; it also assigns weights to the local interaction features of three targets, and after deep integration, obtains a fused interaction feature of one target, thus completing the processing of the feature fusion module.

[0134] Example 2: Input features: 3 local ontology features of the target and 2 local interaction features of the target; The primary feature fusion submodule is implemented using "average pooling fusion". It performs average pooling on the local ontology features of the three targets to obtain one primary fused ontology feature; and performs average pooling on the local interaction features of the two targets to obtain one primary fused interaction feature. Advanced Feature Fusion Submodule: Implemented using "residual fusion + nonlinear transformation", it performs residual connection fusion of three target local ontology features to eliminate redundancy, and then optimizes the feature expression through nonlinear transformation to obtain one target fused ontology feature; the same processing is performed on two target local interaction features to obtain one target fused interaction feature, thus completing the feature fusion module processing.

[0135] In the embodiments of this specification, by splitting the feature fusion module into two sub-modules, namely primary and advanced, a hierarchical processing of feature fusion, namely "basic integration - deep optimization", is realized. The primary sub-module can quickly complete the basic summary of local features, ensuring the comprehensiveness of the fused features and avoiding the omission of core local features. The advanced sub-module can perform deep optimization of local features, combine feature correlation to allocate weights and eliminate redundancy, solve the problem that a single fusion method is difficult to balance comprehensiveness and accuracy, and greatly improve the quality and discriminability of fused features.

[0136] In one optional implementation of this embodiment, the primary feature fusion submodule includes a primary feature fusion unit and a primary nonlinear mapping unit; Based on at least one target local ontology feature and at least one target local interaction feature, the corresponding primary feature fusion submodule obtains primary fused ontology features in the ontology attribute dimension and primary fused interaction features in the interaction attribute dimension, including: Based on at least one target local ontology feature and at least one target local interaction feature, the original fused ontology feature in the ontology attribute dimension and the original fused interaction feature in the interaction attribute dimension are obtained through the corresponding primary feature fusion unit. Based on the original fused ontology features and the original fused interaction features, the primary fused ontology features under the ontology attribute dimension and the primary fused interaction features under the interaction attribute dimension are obtained through the corresponding primary nonlinear mapping units.

[0137] Primary Feature Fusion Unit: The core sub-unit of the primary feature fusion submodule. Its core function is to directly integrate the input target local features (target local ontology features, target local interaction features), complete the feature splicing and summarization, and output the original fused features without nonlinear optimization. It is the foundation of primary fusion. Implementation examples: Example 1: Using a feature splicing unit, multiple sets of local features are spliced ​​in sequence to obtain the original fused features; Example 2: Using a fully connected layer for linear weighting.

[0138] Primary Nonlinear Mapping Unit: An auxiliary sub-unit of the primary feature fusion sub-module, connecting to the primary feature fusion unit. Its core function is to perform nonlinear transformation and optimization on the original fused features output by the primary feature fusion unit, eliminating linear redundancy of the original fused features, enhancing the nonlinear expressive power of the features, and converting the original fused features into primary fused features that meet the requirements of advanced fusion. Implementation examples: Example 1: Using the ReLU activation function to perform nonlinear mapping on the original fused features and filter out invalid features; Example 2: Using the ReLU activation function + lightweight fully connected layer to further optimize the distribution of the original fused features.

[0139] Original fused ontology features: These are fused features obtained by the primary feature fusion unit after preliminary integration of the target local ontology features. They have not undergone nonlinear optimization and only complete feature summarization. They have problems such as linear redundancy and insufficient expression and need further optimization.

[0140] Original fusion interaction features: These are fusion features obtained by the primary feature fusion unit after preliminary integration of the target local interaction features. They have not undergone nonlinear optimization and only complete feature summarization. They need to be optimized by nonlinear mapping before they can be used as primary fusion features.

[0141] The specific process of this scheme is as follows: First, the target local ontology features and target local interaction features output by the feature extraction module are input into the primary feature fusion submodule; then, the two types of target local features are initially integrated by the primary feature fusion unit, and multiple sets of local features are summarized by methods such as splicing and element-wise addition, to obtain the original fused ontology features under the ontology attribute dimension and the original fused interaction features under the interaction attribute dimension; finally, the original fused ontology features and the original fused interaction features are input into the primary nonlinear mapping unit, and through the nonlinear transformation and optimization processing of this unit, the linear redundancy of the original fused features is eliminated, the nonlinear expression of the features is strengthened, and invalid information is filtered out, finally obtaining the primary fused ontology features and the primary fused interaction features.

[0142] In the embodiments of this specification, the primary feature fusion unit focuses on the basic summarization of local features to ensure the comprehensiveness of features and avoid the omission of core local features. At the same time, it adopts a simple and efficient integration method to ensure the efficiency of primary fusion. The primary nonlinear mapping unit effectively eliminates the linear redundancy of the original fused features and enhances the nonlinear expressive ability of the features through nonlinear transformation. This solves the problems of insufficient expression and low discriminative power of the original fused features, making the primary fused features more suitable for the deep optimization needs of the advanced feature fusion submodule.

[0143] In one optional implementation of this embodiment, the advanced feature fusion submodule includes an advanced feature fusion unit and an advanced nonlinear mapping unit; Based on at least one target local ontology feature and at least one target local interaction feature, the corresponding high-level feature fusion submodule obtains the target fused ontology feature in the ontology attribute dimension and the target fused interaction feature in the interaction attribute dimension, including: Based on at least one target local ontology feature and at least one target local interaction feature, the initial fused ontology feature in the ontology attribute dimension and the initial fused interaction feature in the interaction attribute dimension are obtained through the corresponding high-level feature fusion unit. Based on the initial fused ontology features and initial fused interaction features, the target fused ontology features under the ontology attribute dimension and the target fused interaction features under the interaction attribute dimension are obtained through the corresponding high-level nonlinear mapping unit.

[0144] Advanced Feature Fusion Unit: The core sub-unit of the advanced feature fusion submodule. Its core function is to deeply integrate the input target local features (target local ontology features and target local interaction features), and to perform weight allocation and redundancy elimination based on the correlation and importance of the features. The output is an initial fused feature without deep nonlinear optimization, which is the core link of advanced fusion. Implementation examples: Example 1: Using an attention mechanism unit, different weights are assigned to multiple groups of local features to highlight the core features before fusion; Example 2: Using a cross-fusion unit, the correlation between different local features is explored to achieve deep integration.

[0145] Advanced Nonlinear Mapping Unit: An auxiliary sub-unit of the advanced feature fusion sub-module, connecting to the advanced feature fusion unit. Its core function is to perform deep nonlinear transformation and optimization on the initial fused features output by the advanced feature fusion unit, further enhancing the nonlinear expressive power of the features, adjusting the feature distribution, eliminating residual redundancy, and converting the initial fused features into the final usable target fused features. Implementation examples: Example 1: Using the Sigmoid activation function + fully connected layer to deeply optimize the initial fused features and ensure the normalization of the feature vectors; Example 2: Using the Tanh activation function + normalization layer to optimize the feature distribution and improve the stability and discriminative power of the features.

[0146] Initial fused ontology features: These are fused features obtained by the advanced feature fusion unit after deep integration of the target local ontology features. They have not undergone deep nonlinear optimization and have completed weight allocation and redundancy elimination. They are core comprehensive features with ontology attributes and require further deep optimization.

[0147] Initial fusion interaction features: These are fusion features obtained by the advanced feature fusion unit after deep integration of the target local interaction features. They have not undergone deep nonlinear optimization and have completed weight allocation and redundancy elimination. They need to be optimized by deep nonlinear mapping before they can be used as the final target fusion features.

[0148] The specific process of this scheme is as follows: First, the target local ontology features and target local interaction features output by the feature extraction module are input into the high-level feature fusion submodule. Then, the two types of target local features are deeply integrated by the high-level feature fusion unit. Attention mechanisms, cross-fusion, and other methods are used to assign weights based on the correlation and importance of features, eliminating redundant features, and obtaining the initial fused ontology features under the ontology attribute dimension and the initial fused interaction features under the interaction attribute dimension, respectively. Finally, the initial fused ontology features and the initial fused interaction features are input into the high-level nonlinear mapping unit. Through the deep nonlinear transformation and optimization processing of this unit, the nonlinear expressive power of the features is further enhanced, the feature distribution is adjusted, and the remaining redundancy is eliminated, finally obtaining the target fused ontology features and the target fused interaction features, which serve as the direct input for the subsequent output module to generate feature vectors.

[0149] In the embodiments of this specification, the advanced feature fusion submodule is further divided into an advanced feature fusion unit and an advanced nonlinear mapping unit, constructing a complete optimization link of "local features - original fused features - primary fused features - initial fused features - target fused features", which further improves the quality and accuracy of fused features. The advanced feature fusion unit focuses on the deep integration of target local features, highlighting core features and eliminating redundancy through weight allocation, correlation mining, and other methods, solving the problem that primary fused features are difficult to reflect the differences in feature importance, and ensuring the targeting of fused features. The advanced nonlinear mapping unit further enhances the nonlinear expressive power of features, adjusts feature distribution, and eliminates residual redundancy through deep nonlinear transformation and optimization, making the target fused features more in line with the subsequent processing needs of the model, with higher discriminativeness and stability, thereby greatly improving the comprehensiveness, accuracy, and standardization of fused features, providing solid support for the subsequent output module to generate high-quality feature vectors, and further improving the processing effect of the entire feature analysis layer.

[0150] Step 106: Through the fusion layer in the object evaluation model, the target ontology feature vector and the target interaction feature vector are fused to obtain the target value evaluation information of the target object.

[0151] Fusion Layer: In the object evaluation model, the core functional layer that connects the feature analysis layer and the final evaluation output is specifically used to perform hierarchical integration and value calculation of the target ontology feature vector and the target interaction feature vector. It is the core module for realizing object value evaluation.

[0152] Target value assessment information: The final output of the model is a quantitative / grading determination of the comprehensive value of the target object, providing a direct basis for business decisions.

[0153] In one optional implementation of this embodiment, the target ontology feature vector and the target interaction feature vector are fused through a fusion layer in the object evaluation model to obtain the target value evaluation information of the target object, including: The target ontology feature vectors are fused to obtain target ontology features, and the target interaction feature vectors are fused to obtain target interaction features; Based on the set fusion weights, the target local features and target interaction features are fused to obtain the target fused features; Based on the target fusion characteristics and the established nonlinear mapping relationship, target value assessment information is determined.

[0154] Target ontology features: After internal fusion of the target ontology feature vectors, the resulting comprehensive features represent the ontological attributes of the object and are the final feature expression of the object's inherent attributes.

[0155] Target interaction features: After internal fusion of the target interaction feature vector, the resulting comprehensive features represent the interaction attributes of the object and are the final feature expression of the object's social, influence, and propagation interaction behaviors.

[0156] Fusion weight: A proportional parameter set manually or adaptively by the model to balance the importance of target ontology features and target interaction features in the final evaluation, and can be flexibly adjusted according to business needs.

[0157] Target fusion feature: The final comprehensive feature obtained by weighted fusion of target ontology features and target interaction features according to the set fusion weights, which fully covers all core information of the object's ontology and interaction.

[0158] Nonlinear mapping relationship: Pre-defined nonlinear transformation rules (such as activation functions and fully connected mappings) are used to transform high-dimensional target fusion features into intuitive and quantifiable value evaluation results.

[0159] This solution is the final output stage of the object evaluation model, forming a complete, progressive logic with the previous feature extraction and feature analysis layers. Its core is the hierarchical fusion and value transformation of two-dimensional feature vectors through a fusion layer. The specific process consists of three steps: First, within the fusion layer, the target ontology feature vector and target interaction feature vector are independently fused, eliminating redundant information to obtain highly condensed target ontology features and target interaction features. Second, fusion weights are set according to business needs, and the two independent comprehensive features are weighted and fused to obtain target fusion features that take into account both ontology and interaction attributes. Third, based on a preset nonlinear mapping relationship, the high-dimensional target fusion features are converted into standardized evaluation results, ultimately outputting the target object's target value evaluation information. The entire process achieves a closed loop of "feature vector → hierarchical fusion → weighted integration → value output," completing the transformation from feature data to business value.

[0160] Taking game user value assessment as an example: First, the target ontology feature vector (including level, combat power, online time) and target interaction feature vector (including object radiation feature vector, object-driven feature vector, and positive information feature vector) output by the previous feature analysis layer are input into the fusion layer. The two types of vectors are internally fused to obtain the condensed target ontology features and target interaction features. Second, the fusion weights are set: ontology feature weight 0.4 and interaction feature weight 0.6 (emphasizing interaction value). The two types of features are weighted and fused to obtain the target fusion feature. Third, the target fusion feature is mapped to a value score of 0-100 through the ReLU nonlinear mapping function. Finally, the target value assessment information of the game user is output (e.g., 90 points: high-value user, 60 points: medium-value user).

[0161] In the embodiments of this specification, a scientifically rigorous value assessment logic is constructed through step-by-step processing of hierarchical fusion, weighted integration, and nonlinear mapping. First, the independent fusion of dual-dimensional feature vectors ensures the integrity of the ontology and interaction features, avoiding mutual interference between features of different dimensions. The fusion method based on set weights can flexibly adapt to the assessment emphasis of different business scenarios, greatly improving the model's versatility and adjustability. The nonlinear mapping relationship solves the problem that high-dimensional features cannot be directly applied, making the assessment results more accurate and more in line with actual business needs. At the same time, this solution is perfectly integrated with the previous modular design of feature extraction and feature fusion, forming a standardized processing system for the entire process from raw parameters to value assessment. This not only improves the accuracy and stability of value assessment, but also simplifies the cost of model optimization and iteration, providing reliable support for the accurate assessment and classification management of target objects.

[0162] In another optional implementation of this embodiment, the target value assessment information of the target object is obtained by fusing the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model. This includes: directly concatenating the target ontology feature vector and the target interaction feature vector to obtain initial concatenated features; dynamically weighting and optimizing the initial concatenated features through the adaptive attention unit of the fusion layer to extract core related features and obtain adaptive optimized features; performing global feature condensation on the adaptive optimized features to obtain target fusion features; and determining the target value assessment information based on the target fusion features and the set nonlinear mapping relationship.

[0163] In an optional implementation of this embodiment, the above object evaluation method further includes: Based on the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension, the object feature description information of the target object is determined, and based on the object feature description information, operational suggestions for the target object are generated.

[0164] Object feature description information: Based on the target ontology feature vector and the target interaction feature vector, the concrete and understandable feature summary information is extracted and transformed. It is not just abstract vector data, but a textual / quantitative description that can clearly reflect the core characteristics of the target object, and fully covers ontology attributes and interaction attributes. Operational recommendations: Based on object feature descriptions and combined with business scenario requirements (such as game operation), targeted and actionable operational recommendations are generated to guide the precise operation of target objects, transforming model evaluation results into actual business value and adapting to target objects with different feature types.

[0165] This solution supplements and extends the functionality of the object evaluation model described earlier. Building upon the initial target value assessment, it further explores the business value of dual-dimensional feature vectors, forming a complete closed loop of "feature extraction → feature fusion → value assessment → feature description → operational suggestions." The specific process is as follows: First, it calls upon the previously generated target ontology feature vectors (representing the object's inherent basic attributes) and target interaction feature vectors (representing interactive behaviors such as object radiation, object-driven behavior, and positive propagation). Second, it analyzes and refines these two types of feature vectors, transforming abstract vector data into concrete object feature descriptions, clearly presenting the core characteristics of the target object (such as basic capabilities at the ontology level and influence at the interaction level). Finally, based on the refined object feature descriptions and combined with the operational goals of specific business scenarios (such as retention, activation, and conversion), it generates targeted operational suggestions tailored to the characteristics of the target object, realizing the transformation from "evaluation data" to "business action." This solution complements and parallels the previously described fusion layer value assessment solution, outputting both quantitative value assessment information and concrete feature descriptions and actionable operational guidance, thus perfecting the entire object evaluation system.

[0166] For example, the target ontology feature vector generated above (including character level 80, combat power 50,000, average daily online time 3 hours) and target interaction feature vector (including social circle size of 40 people, leading core dungeons 12 times / month, external traffic of 8 people / month). Determine the object's characteristic description information: Analyze and refine the two types of vectors to generate a concrete description - "The user of this game has a good basic foundation (level 80, combat power is above average, and daily online stability), outstanding interaction ability (active social circle, strong ability to lead dungeons and attract external traffic), and belongs to the above-average value user"; Based on the above characteristics and combined with the game's operational goals (improving high-value user retention and strengthening user acquisition capabilities), the following targeted suggestions are generated: "1. Push exclusive rewards for leading dungeons to encourage the user to continue leading teams and attract new players; 2. Open an exclusive referral invitation channel for the user and give them extra rewards to stimulate their enthusiasm for referrals; 3. Invite the user to join the game's core player community to increase user stickiness and leverage their social influence."

[0167] In the embodiments of this specification, the object feature description information concretizes abstract vector data, making it easier for operators to quickly grasp the core characteristics of the target object and solving the problem that feature vectors are difficult to interpret directly. The operational suggestions generated based on the feature description are highly targeted, avoiding a "one-size-fits-all" operational model, and can accurately match the characteristics of the target object, improving operational efficiency and effectiveness. At the same time, the solution is perfectly integrated with the feature extraction, feature fusion, and value assessment processes described above, forming a complete closed loop of "data collection → feature processing → assessment and analysis → business implementation," further enhancing the practicality and commercial value of the object assessment model, so that the model can not only "assess value" but also "guide operations," providing more comprehensive and practical support for business decisions.

[0168] The following section uses a game scenario as an example to describe in detail the feature analysis process of the embodiments in this specification.

[0169] First, the embodiments in this specification do not use a single deep fully connected network (DNN) or gradient boosting tree (such as XGBoost), but instead construct a hybrid architecture of "four-branch neural network + configurable attention fusion module". The design considerations mainly include the following three aspects: (1) Adapting to complex nonlinear interaction features: User influence is not a simple linear superposition of features. There is a synergistic amplification effect between features of different dimensions. For example, when the identity of "guild leader" (social dimension feature) and "100% core dungeon completion rate" (gameplay dimension feature) coexist, the user influence will show nonlinear enhancement. Neural networks have a natural ability to learn complex nonlinear patterns and can effectively capture such cross-dimensional feature interaction relationships.

[0170] (2) Respecting business grouping to improve learning efficiency and interpretability: The input parameters of this solution come from four independent business concepts: personal foundation, social linkage, gameplay-driven, and word-of-mouth dissemination. If multi-dimensional features are mixed and input into a single network, it is easy for the model to focus on the deep patterns of each dimension. By setting up a four-branch structure, each branch is dedicated to learning the feature patterns of the corresponding dimension (such as the social branch focusing on mining the correlation between "guild position level" and "social circle size"). Analogous to the multi-expert collaborative consultation model, the feature learning efficiency is higher than that of a single general model, and the intermediate output results of each branch have direct business interpretability.

[0171] (3) Support dynamic scene adaptation: The focus of game operation changes dynamically with version iteration and event cycle. For example, the evaluation of "gameplay-driven" is emphasized during the launch of new expansion packs, while the evaluation of "social organization" is emphasized during guild wars. The configurable attention fusion module supports the dynamic adjustment of the contribution weight of the four branches through external configuration parameters (such as weight configuration dictionary). The evaluation focus can be switched without retraining the model, which significantly improves the operational adaptation flexibility of the model.

[0172] The embodiments in this specification adopt a four-branch independent architecture (four feature analysis layers). Each branch model independently processes the attribute features of the corresponding dimension and extracts the core features, which can be specifically adapted to the business characteristics and data distribution patterns of different dimension attributes to ensure the accuracy and effectiveness of feature learning.

[0173] See Figure 2 , Figure 2 A schematic diagram of the structure of a feature analysis layer in a model provided according to an embodiment of this specification is shown.

[0174] like Figure 2 As shown, Figure 2 This demonstrates the inference process of a single-dimensional feature analysis layer, which extracts deep features from the original input features and outputs standardized evaluation results. The functions of each module are as follows: Input module: Receives the original feature vectors under the corresponding business dimension as the initial data input for model processing.

[0175] Convolutional layer: By performing local window scanning on the input features through linear weighted summation, it extracts local correlation patterns between features and captures the basic local features contained in the original features.

[0176] Activation function module: Performs nonlinear transformation on local features output by convolutional layers, filters out invalid negative features, enhances the expressive power of effective features, and provides nonlinear discriminative power for subsequent feature fusion.

[0177] Flattening module: Converts the two-dimensional feature map obtained after processing by the convolutional layer and activation function into a one-dimensional vector form to adapt to the input format requirements of the subsequent fully connected layer.

[0178] Fully connected layer 1: Performs linear combination on the flattened one-dimensional local feature vectors to achieve global feature fusion, integrating scattered local features into global features that can reflect the overall pattern of the input features.

[0179] Activation function module: performs non-linear filtering on the global features output by fully connected layer 1 to further remove redundant information and enhance the expression of core features.

[0180] Fully connected layer 2: The global features after nonlinear filtering are linearly combined again to extract more representative high-level core features, compress feature dimensions and focus on key information.

[0181] Activation function module: Performs final nonlinear processing on the high-level features output by fully connected layer 2 to ensure that the output features meet the numerical requirements of subsequent mapping.

[0182] Output layer (fully connected): Maps high-level core features to a preset target dimension to generate standardized feature vectors aligned with business objectives.

[0183] Final output module: Outputs the evaluation score vector corresponding to the business dimension, completing the feature extraction and evaluation under that dimension.

[0184] It should be noted that the size of the convolutional kernel, the number of neurons in the fully connected layer, and the choice of activation function in the model are all closely related to the properties of the input parameters, including the number of dimensions of the input features, business characteristics, data distribution patterns, and the complexity of the relationships between features.

[0185] Kernel size: Determined based on the number of dimensions of the input features and the coverage of local associations. The more dimensions and the more complex the interactions between features, the larger the kernel is usually chosen to cover more interaction patterns between features; when the dimensions are fewer and the associations are simpler, a smaller kernel is chosen to accurately capture local patterns.

[0186] Number of neurons in the fully connected layer: The number of neurons is determined based on the complexity of the input features and the granularity of the business objectives. The more complex the feature interactions and the more refined the business evaluation dimensions, the more neurons are set to carry richer feature expressions; conversely, the number is reduced appropriately to avoid redundant calculations.

[0187] Activation function: Select the appropriate activation function type based on the distribution characteristics of the input features and the numerical range requirements of the business output. For example, when it is necessary to map features to a specific range, retain weak negative signals, or strengthen positive features, select the appropriate activation function type to match the expression requirements of features of different dimensions.

[0188] Based on the above general process, and considering the feature analysis requirements of the four business dimensions in the game scenario, adapted model parameters were designed as follows: Feature analysis layer of user basic attribute dimension: This feature analysis layer focuses on analyzing a player's 5-dimensional personal basic data, ultimately outputting an assessment result of "user's basic radiation capability". Combining the characteristics of fewer input dimensions and stronger linear correlations, a 3×3 convolutional kernel is set to accurately capture the local linkages between basic attributes; the number of neurons in the fully connected layer is 64 and 32, adapting to the complexity of the basic features; and the ReLU activation function is used to filter out invalid negative features and strengthen the expression of positive basic capabilities.

[0189] Feature analysis layer of social interaction attribute dimension: This feature analysis layer focuses on analyzing player-level social interaction data, ultimately outputting an evaluation result of "group reach and quality". Considering the characteristics of multiple input dimensions and complex social behavioral interactions, a 5×5 convolutional kernel is set to cover more correlation patterns between social features; the number of neurons in the fully connected layers is 128 and 64, carrying richer expressions of social behavioral features; and the LeakyReLU activation function is selected to preserve weak negative social signals and adapt to subtle differences in social relationships.

[0190] Gameplay-driven feature analysis layer based on attribute dimensions: This feature analysis layer focuses on analyzing the five-dimensional gameplay-driven data of players, ultimately outputting an evaluation result of "group behavior driving ability". Combining the characteristics of moderate input dimensions and strong gameplay behavior coordination, a 4×4 convolutional kernel is set to capture the collaborative relationships between gameplay behaviors such as leading a team and clearing levels; the number of neurons in the fully connected layers is 128 and 64, adapting to the complexity of gameplay coordination features; and the ReLU activation function is selected to strengthen the expression of positive gameplay-driven signals.

[0191] Feature analysis layer of word-of-mouth communication attribute dimensions: This feature analysis layer focuses on analyzing players' 4D word-of-mouth dissemination data, ultimately outputting an evaluation result of "influence amplification ability". Considering the characteristics of limited input dimensions and the probabilistic nature of word-of-mouth features, a 3×3 convolutional kernel is set to accurately capture the local correlations between word-of-mouth dissemination features; the number of neurons in the fully connected layers is 64 and 32, adapting to the complexity of word-of-mouth features; and the Sigmoid activation function is selected to map features to the 0-1 range, matching the probabilistic expression requirements of word-of-mouth.

[0192] For example, the specific processing procedure of the above feature analysis layer is as follows: Input the feature vector (a combination of object parameters) of the player's basic user attributes: [P1:0.8,P2:0.9,P3:0.7,P4:0.8,P5:0.6].

[0193] Convolutional layer: Convolutional kernel settings: 3×3 (adapts to 5-dimensional input, sliding window covers 3 consecutive features, captures local correlations), randomly initialized weights [0.2, -0.5, 0.3], bias term j=0.1.

[0194] Sliding window calculation: Window 1 (P1-P3): 0.8×0.2+0.9×(-0.5)+0.7×0.3+0.1=0.02; Window 2 (P2-P4): 0.9×0.2+0.7×(-0.5)+0.8×0.3+0.1=0.17; Window 3 (P3-P5): 0.7×0.2+0.8×(-0.5)+0.6×0.3+0.1=0.02.

[0195] Output: 3 local feature values ​​[0.02, 0.17, 0.02] (output of a single convolution kernel; the actual model has 64 convolution kernels, outputting a total of 64 × 3 = 192 local feature signals).

[0196] ReLU activation function: performs non-linear processing on each value of the convolutional layer output: if the value is ≥0, it is retained; if it is <0, it is set to 0.

[0197] Example: If a convolution kernel outputs [-0.3, 0.4, -0.1], after ReLU it becomes [0.0, 0.4, 0.0], filtering out invalid negative features and strengthening the expression of positive features.

[0198] Flatten (convert feature map into one-dimensional vector) The two-dimensional feature map (64 convolutional kernels × 3 feature values) after convolution + ReLU is flattened into a one-dimensional vector: [x1,x2,...,x192] (192 elements in total), which prepares for the subsequent fully connected layer.

[0199] Fully connected layer 1: Neuron configuration: 64.

[0200] Calculation logic: For the 192 flattened local features, perform a linear weighted summation using weights [w1, w2, ..., w192], and add a bias term to obtain 64 preliminary global features: z_i=w1×x1+w2×x2+...+w192×x192+j(i=1~64) Function: Integrates local features into global features to capture the overall pattern of an individual's basic attributes. ReLU activation function (non-linear filtering): The output of fully connected layer 1 is processed non-linearly again to filter out invalid negative features and enhance the discriminative power of effective features.

[0201] Fully connected layer 2: Neuron configuration: 32.

[0202] Computational logic: The 64 outputs of the fully connected layer 1 are linearly weighted and summed to obtain 32 high-level features.

[0203] Function: To remove redundant information and extract the core characteristics that best represent an individual's basic abilities.

[0204] ReLU activation function (nonlinear filtering): The final nonlinear processing ensures that the output features are non-negative, meeting the numerical requirements for subsequent fusion.

[0205] Output layer (fully connected | mapped to target dimension): Maps 32 high-level features to a 5-dimensional personal base feature vector (aligned with the input dimension for easy business interpretation): Example output: z1=[0.667,0.5,0.3,0.4,0.6] The processing flow for social interaction, gameplay-driven, and word-of-mouth marketing branches is completely consistent with the basic personal branch, with only the convolutional kernel size, number of neurons, and activation function adjusted according to their own feature dimensions and business characteristics: Social interaction branch: Input 6-dimensional features, 5×5 convolutional kernel, 128→64 fully connected layers, LeakyReLU activation function, and finally output 6-dimensional vector z2=[0.8,0.75,0.9,0.85,0.7,0.8].

[0206] Gameplay-driven branching: Input 5-dimensional features, 4×4 convolutional kernels, 128→64 fully connected layers, ReLU activation function, and finally output a 5-dimensional vector z3=[0.9,0.85,0.95,0.8,0.9].

[0207] Word-of-mouth propagation branch: Input 4-dimensional features, 3×3 convolutional kernel, 64→32 fully connected layers, Sigmoid activation function, and finally output 4-dimensional vector z4=[0.7,0.65,0.6,0.75].

[0208] Fusion layer: Configurable attention-weighted + Sigmoid scaling Branch vector summation: Summing the output vectors of each branch yields the overall feature strength of that branch. User base: sum(z1) = 0.667 + 0.5 + 0.3 + 0.4 + 0.6 = 2.467; Social interaction: sum(z2) = 0.8 + 0.75 + 0.9 + 0.85 + 0.7 + 0.8 = 4.8; Gameplay-driven: sum(z3) = 0.9 + 0.85 + 0.95 + 0.8 + 0.9 = 4.4; Word-of-mouth marketing: sum(z4) = 0.7 + 0.65 + 0.6 + 0.75 = 2.7; Configurable attention weighting (adapted to operational scenarios): Set weights based on the game's operational focus (example: new expansion launch, focusing on gameplay and social interaction): personal base 10%, social interaction 40%, gameplay-driven 40%, word-of-mouth spread 10%.

[0209] Calculate the weighted total score S_branch: S_branch=2.467×0.1+4.8×0.4+4.4×0.4+2.7×0.1=0.2467+1.92+1.76+0.27=4.1967.

[0210] Sigmoid scaling (mapped to 0-100 points): The S_branch is mapped to the 0-1 range using the Sigmoid function, and then scaled up by 100 times to obtain the final influence score. S_sigmoid=1 / (1+e^(-4.1967))≈0.985.

[0211] S_final = 0.985 × 100 ≈ 98.5.

[0212] Based on the influence score (or value score) mentioned above, the influence level of the object is determined. This can be determined based on the following mapping table—Table (1): Table 1

[0213] Influence type determination (based on branch weighted percentage): Calculate the weighted score of each branch as a percentage of the total score, and determine the type based on the relative magnitude: The gameplay-driven branch weighted score is 4.4 × 0.4 = 1.76 (accounting for 1.76 / 4.1967 ≈ 41.9%). Weighted score for social interaction branch: 4.8 × 0.4 = 1.92 (accounting for 1.92 / 4.1967 ≈ 45.8%). Individual basic branch weighted score: 2.467 × 0.1 = 0.2467 (percentage ≈ 5.9%); Weighted score for word-of-mouth marketing: 2.7 × 0.1 = 0.27 (percentage ≈ 6.4%). Meets the criteria for social core type: Social weighted score (1.92) ≥ the maximum score of the other three categories (gameplay 1.76) × 1.2 → 1.92 ≥ 2.112? This does not satisfy the condition. In the actual example, if the weights are adjusted to 50% for social and 30% for gameplay, then it satisfies the condition. This explanation uses the "gameplay core type" logic as an example. If the weighted score of the gameplay is greater than or equal to 1.2 times the maximum score of the other three categories, and accounts for more than 45% of the total score, it is determined to be a core gameplay type. Core Feature Profile Generation: Automatically fill in templates based on branch behavior and type.

[0214] "S-level gameplay core type" players have a solid personal foundation (level 80, above-average combat power, stable daily online time), and their core value lies in top-level dungeon command and teaching ability (high frequency of leading teams in core dungeons, outstanding clearance rate). They also have excellent guild organization and social cohesion, and have a positive reputation both inside and outside the game. They are the dual core benchmark of server gameplay and community.

[0215] Operational suggestion generation: Based on the type and shortcomings, generate actionable operational actions.

[0216] Grant the title of "Server Chief Commander", which will be displayed on the in-game loading screen and in the guild channel to strengthen their authoritative identity in the gameplay.

[0217] Invite to serve as the core commander of the new version's开荒团 (it seems there is a wrong word here, maybe it should be "raid group"), and provide resource support such as exclusive team fashion and dungeon acceleration items.

[0218] Guide them to carry out novice dungeon teaching activities, and give rewards such as additional experience bonuses and bound diamonds to improve the retention rate of new players.

[0219] Invite to join the core player think tank, participate in the discussion of gameplay optimization, and enhance their sense of belonging and voice.

[0220] Regularly push exclusive benefits for high-value players (such as limited appearances and priority experience qualifications), and focus on maintaining to avoid loss.

[0221] Output structured JSON data: json { "Player ID": "Fengwu Tianxia", "Comprehensive Influence Score": 98.5, "Influence Level": "S", "Influence Type": "Gameplay Core Type", "Core Feature Portrait": "Top-level Dungeon Command and Teaching Ability (extremely strong gameplay driving ability)", "Excellent Guild Organization Ability and Social Cohesion", "Solid personal foundation, good payment and activity", "Positive reputation inside and outside the game" , "Operation Suggestions": "Can grant the title of 'Server Chief Commander'", "Invite to serve as the core commander of the new version's raid group", "Provide exclusive team fashion to strengthen their team identity" }。

[0222] The following combines with the attached Figure 3 , taking the application of the object evaluation method provided in this specification in the evaluation of user value in the game as an example, to further illustrate the object evaluation method. Among them, Figure 3 Figur shows the processing procedure flowchart of an object evaluation method provided in an embodiment of this specification, which specifically includes the following steps.

[0223] It should be noted that there may be some inaccuracies in the translation due to possible unclear or incorrect expressions in the original text. For example, the term "开荒团" is not a common and accurately translated word, and it is tentatively translated as "raid group" here. You may need to further confirm and correct according to the actual situation.Step 302: Obtain the object parameters of the target object to be evaluated under the dimensions of user basic attributes, social interaction attributes, gameplay-driven attributes, and word-of-mouth dissemination attributes.

[0224] Step 304: Based on the object parameters under the user basic attribute dimension, obtain the user feature vector under the user basic attribute dimension through the feature analysis layer corresponding to the user basic attribute dimension in the object evaluation model.

[0225] Step 306: Based on the object parameters under the social linkage attribute dimension, obtain the social feature vector under the social linkage attribute dimension through the feature analysis layer corresponding to the social linkage attribute dimension in the object evaluation model.

[0226] Step 308: Based on the object parameters under the gameplay-driven attribute dimension, obtain the gameplay-driven feature vector under the gameplay-driven attribute dimension through the feature analysis layer corresponding to the gameplay-driven attribute dimension in the object evaluation model.

[0227] Step 310: Based on the object parameters under the word-of-mouth propagation attribute dimension, obtain the word-of-mouth propagation feature vector under the word-of-mouth propagation attribute dimension through the feature analysis layer corresponding to the word-of-mouth propagation attribute dimension in the object evaluation model.

[0228] Step 312: Through the fusion layer in the object evaluation model, the user feature vector, social feature vector, gameplay-driven feature vector, and word-of-mouth spread feature vector are fused to obtain the target value evaluation information of the target object.

[0229] Step 314: Based on user feature vectors, social feature vectors, gameplay-driven feature vectors, and word-of-mouth propagation feature vectors, determine the object feature description information of the target object, and generate operational suggestions for the target object based on the object feature description information.

[0230] It should be noted that the object evaluation method provided in this manual can be applied to multiple industries or scenarios, such as virtual reality processing software, home entertainment product software, digital cultural product production software, digital cultural creative software, digital cultural creative design, education, news, cultural content industry software, digital publishing software, digital music development and production, and digital mobile multimedia development and production. In some cases, it can also be applied to fields such as animation and game production engine software and development systems, game and animation software, animation and game digital content services, digital film and television development and production, and digital performance development and production.

[0231] Corresponding to the above method embodiments, this application also provides an object evaluation apparatus embodiment. Figure 4 A schematic diagram of an object evaluation device according to an embodiment of this application is shown. Figure 4 As shown, the device includes: The acquisition module 402 is configured to acquire object parameters of the target object to be evaluated in the ontology attribute dimension and the interaction attribute dimension. The feature determination module 404 is configured to obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension based on the object parameters under the ontology attribute dimension and the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model. The feature analysis layer is based on the attribute determination of the object parameters under the corresponding attribute dimension. The object value determination module 406 is configured to fuse the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object.

[0232] Optionally, the interaction attribute dimension includes an object radiation sub-dimension, an object driving sub-dimension, and a positive information diffusion sub-dimension. The object radiation sub-dimension characterizes the range of other objects that the target object can influence, the object driving sub-dimension characterizes the effect of the target object on increasing the activity of other objects, and the positive information diffusion sub-dimension characterizes the effect of the target object on the propagation of positive information. The feature determination module 404 is further configured as follows: Based on the object parameters under the ontology attribute dimension, the target ontology feature vector under the ontology attribute dimension is obtained through the feature analysis layer corresponding to the ontology attribute dimension. Based on the object parameters under the object radiation sub-dimension, object driving sub-dimension, and positive information diffusion sub-dimension, the target interaction feature vector under the interaction attribute dimension is obtained through the feature analysis layer corresponding to the interaction attribute dimension. In the object evaluation model, one feature analysis layer is used to process the object parameters of at least one sub-dimension under the interaction attribute dimension.

[0233] Optionally, a feature analysis layer in the object evaluation model is used to process object parameters in a sub-dimension of the interaction attribute dimension; the target interaction feature vector includes the object radiation feature vector in the object radiation sub-dimension, the object driving feature vector in the object driving sub-dimension, and the positive information feature vector in the positive information diffusion sub-dimension; the feature determination module 404 is further configured to: Based on the object parameters under the object radiation sub-dimension, the object radiation feature vector under the object radiation sub-dimension is obtained through the feature analysis layer corresponding to the object radiation sub-dimension. Based on the object parameters under the object-driven sub-dimension, the object-driven feature vector under the object-driven sub-dimension is obtained through the feature analysis layer corresponding to the object-driven sub-dimension. Based on the object parameters under the positive information diffusion sub-dimension, the positive information feature vector under the positive information diffusion sub-dimension is obtained through the feature analysis layer corresponding to the positive information diffusion sub-dimension.

[0234] Optionally, the feature analysis layer includes a feature extraction module, a feature fusion module, and an output module; the feature determination module 404 is further configured as follows: Based on object parameters under the ontology attribute dimension and interaction attribute dimension, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are extracted through the feature extraction module in the corresponding feature analysis layer. Based on at least one target local ontology feature and at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the feature fusion module in the corresponding feature analysis layer. Based on the fusion of ontology features and fusion of interaction features, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the output module in the corresponding feature analysis layer.

[0235] Optionally, the feature extraction module includes a feature extraction unit and a pre-nonlinear mapping unit; the feature determination module 404 is further configured to: Based on object parameters under the ontology attribute dimension and the interaction attribute dimension, at least one initial local ontology feature under the ontology attribute dimension and at least one initial local interaction feature under the interaction attribute dimension are extracted through the corresponding feature extraction unit. Based on at least one initial local ontology feature and at least one initial local interaction feature, at least one target local ontology feature in the ontology attribute dimension and at least one target local interaction feature in the interaction attribute dimension are obtained through the corresponding pre-nonlinear mapping unit.

[0236] Optionally, the feature fusion module includes a primary feature fusion submodule and a high-level feature fusion submodule; the feature determination module 404 is further configured to: Based on at least one target local ontology feature and at least one target local interaction feature, the primary fused ontology feature under the ontology attribute dimension and the primary fused interaction feature under the interaction attribute dimension are obtained through the corresponding primary feature fusion submodule. Based on at least one target local ontology feature and at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the corresponding high-level feature fusion submodule.

[0237] Optionally, the primary feature fusion submodule includes a primary feature fusion unit and a primary nonlinear mapping unit; the feature determination module 404 is further configured to: Based on at least one target local ontology feature and at least one target local interaction feature, the original fused ontology feature in the ontology attribute dimension and the original fused interaction feature in the interaction attribute dimension are obtained through the corresponding primary feature fusion unit. Based on the original fused ontology features and the original fused interaction features, the primary fused ontology features under the ontology attribute dimension and the primary fused interaction features under the interaction attribute dimension are obtained through the corresponding primary nonlinear mapping units.

[0238] Optionally, the advanced feature fusion submodule includes an advanced feature fusion unit and an advanced nonlinear mapping unit; the feature determination module 404 is further configured to: Based on at least one target local ontology feature and at least one target local interaction feature, the initial fused ontology feature in the ontology attribute dimension and the initial fused interaction feature in the interaction attribute dimension are obtained through the corresponding high-level feature fusion unit. Based on the initial fused ontology features and initial fused interaction features, the target fused ontology features under the ontology attribute dimension and the target fused interaction features under the interaction attribute dimension are obtained through the corresponding high-level nonlinear mapping unit.

[0239] The object value determination module 406 is configured as follows: The target ontology feature vectors are fused to obtain target ontology features, and the target interaction feature vectors are fused to obtain target interaction features; Based on the set fusion weights, the target local features and target interaction features are fused to obtain the target fused features; Based on the target fusion characteristics and the established nonlinear mapping relationship, target value assessment information is determined.

[0240] Optionally, it also includes an operations management module, configured as follows: Based on the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension, the object feature description information of the target object is determined, and based on the object feature description information, operational suggestions for the target object are generated.

[0241] The above is a schematic scheme of an object evaluation device according to this embodiment. It should be noted that the technical solution of this object evaluation device and the technical solution of the object evaluation method described above belong to the same concept. Details not described in detail in the technical solution of the object evaluation device can be found in the description of the technical solution of the object evaluation method described above. Furthermore, the components in the device embodiment should be understood as functional modules necessary to implement each step of the program flow or each step of the method; these functional modules are not actual functional divisions or separations. The device claim defined by such a set of functional modules should be understood as a functional module architecture that primarily implements the solution through the computer program described in the specification, and not as a physical device that primarily implements the solution through hardware.

[0242] Figure 5 A structural block diagram of a computing device 500 according to an embodiment of this application is shown. The components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and a database 550 is used to store data.

[0243] The computing device 500 also includes an access device 540, which enables the computing device 500 to communicate via one or more networks 560. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), a 5G communication network, or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0244] In one embodiment of this application, the aforementioned components of the computing device 500 and Figure 5 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 5 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0245] Computing device 500 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones, in-vehicle computers, POS machines, game consoles, etc.), wearable computing devices (e.g., smartwatches, smart glasses, etc.), smart home appliances, multimedia playback devices, smart voice interaction devices, or other types of mobile devices, or stationary computing devices such as desktop computers or PCs. Computing device 500 can also be a mobile or stationary server. A server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0246] The processor 520 is used to execute computer-executable instructions for the object evaluation method.

[0247] The above is a schematic representation of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the object evaluation method described above belong to the same concept. Details not described in detail in the technical solution of the computing device can be found in the description of the technical solution of the object evaluation method described above.

[0248] An embodiment of this application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, are used for an object evaluation method.

[0249] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the object evaluation method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the object evaluation method described above.

[0250] An embodiment of this application also provides a chip that stores a computer program, which, when executed by the chip, implements the steps of the object evaluation method.

[0251] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the object evaluation method described above.

[0252] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the object evaluation method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the object evaluation method described above.

[0253] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0254] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0255] An embodiment of this application also provides a chip that stores a computer program, which, when executed by the chip, implements the steps of the object evaluation method.

[0256] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0257] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0258] The preferred embodiments disclosed above are merely illustrative of this application. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this application. These embodiments are selected and specifically described in this application to better explain the principles and practical applications of this application, thereby enabling those skilled in the art to better understand and utilize this application. This application is limited only by the claims and their full scope and equivalents.

Claims

1. An object evaluation method, characterized in that, include: Obtain the object parameters of the target object to be evaluated in terms of ontology attributes and interaction attributes; Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the corresponding feature analysis layer in the object evaluation model. The feature analysis layer is determined based on the attributes of the object parameters under the corresponding attribute dimension. The target ontology feature vector and the target interaction feature vector are fused through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object.

2. The method according to claim 1, characterized in that, The interactive attribute dimension includes an object radiation sub-dimension, an object driving sub-dimension, and a positive information diffusion sub-dimension. The object radiation sub-dimension is used to characterize the range of other objects that the target object can influence. The object driving sub-dimension is used to characterize the effect of the target object on improving the activity of other objects. The positive information diffusion sub-dimension is used to characterize the effect of the target object on spreading positive information. The object parameters based on the ontology attribute dimension and the interaction attribute dimension are used to obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model, including: Based on the object parameters under the ontology attribute dimension, the target ontology feature vector under the ontology attribute dimension is obtained through the feature analysis layer corresponding to the ontology attribute dimension. Based on the object parameters under the object radiation sub-dimension, object driving sub-dimension, and positive information diffusion sub-dimension, the target interaction feature vector under the interaction attribute dimension is obtained through the feature analysis layer corresponding to the interaction attribute dimension. In the object evaluation model, one feature analysis layer is used to process the object parameters of at least one sub-dimension under the interaction attribute dimension.

3. The method according to claim 2, characterized in that, In the object evaluation model, a feature analysis layer is used to process the object parameters of a sub-dimension under the interaction attribute dimension. The target interaction feature vector includes the object radiation feature vector under the object radiation sub-dimension, the object driving feature vector under the object driving sub-dimension, and the positive information feature vector under the positive information diffusion sub-dimension. The object parameters based on the object radiation sub-dimension, object driving sub-dimension, and positive information diffusion sub-dimension are used to obtain the target interaction feature vector under the interaction attribute dimension through the feature analysis layer corresponding to the interaction attribute dimension, including: Based on the object parameters under the object radiation sub-dimension, the object radiation feature vector under the object radiation sub-dimension is obtained through the feature analysis layer corresponding to the object radiation sub-dimension. Based on the object parameters under the object-driven sub-dimension, the object-driven feature vector under the object-driven sub-dimension is obtained through the feature analysis layer corresponding to the object-driven sub-dimension. Based on the object parameters under the positive information diffusion sub-dimension, the positive information feature vector under the positive information diffusion sub-dimension is obtained through the feature analysis layer corresponding to the positive information diffusion sub-dimension.

4. The method according to claim 1, characterized in that, The feature analysis layer includes a feature extraction module, a feature fusion module, and an output module; The object parameters based on the ontology attribute dimension and the interaction attribute dimension are used to obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model, including: Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are extracted by the feature extraction module in the corresponding feature analysis layer. Based on the at least one target local ontology feature and the at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the feature fusion module in the corresponding feature analysis layer. Based on the fused ontology features and the fused interaction features, the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension are obtained through the output module in the corresponding feature analysis layer.

5. The method according to claim 4, characterized in that, The feature extraction module includes a feature extraction unit and a pre-nonlinear mapping unit; The object parameters based on the ontology attribute dimension and the interaction attribute dimension are used to extract at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension through the feature extraction module in the corresponding feature analysis layer, including: Based on the object parameters under the ontology attribute dimension and the interaction attribute dimension, at least one initial local ontology feature under the ontology attribute dimension and at least one initial local interaction feature under the interaction attribute dimension are extracted through the corresponding feature extraction unit. Based on the at least one initial local ontology feature and the at least one initial local interaction feature, at least one target local ontology feature under the ontology attribute dimension and at least one target local interaction feature under the interaction attribute dimension are obtained through the corresponding pre-nonlinear mapping unit.

6. The method according to claim 4, characterized in that, The feature fusion module includes a primary feature fusion submodule and a high-level feature fusion submodule; The step of obtaining target fused ontology features under the ontology attribute dimension and target fused interaction features under the interaction attribute dimension based on the at least one target local ontology feature and the at least one target local interaction feature through the feature fusion module in the corresponding feature analysis layer includes: Based on the at least one target local ontology feature and the at least one target local interaction feature, the primary fused ontology feature under the ontology attribute dimension and the primary fused interaction feature under the interaction attribute dimension are obtained through the corresponding primary feature fusion submodule. Based on the at least one target local ontology feature and the at least one target local interaction feature, the target fused ontology feature under the ontology attribute dimension and the target fused interaction feature under the interaction attribute dimension are obtained through the corresponding high-level feature fusion submodule.

7. The method according to claim 6, characterized in that, The primary feature fusion submodule includes a primary feature fusion unit and a primary nonlinear mapping unit; The step of obtaining primary fused ontology features under the ontology attribute dimension and primary fused interaction features under the interaction attribute dimension based on the at least one target local ontology feature and the at least one target local interaction feature through the corresponding primary feature fusion submodule includes: Based on the at least one target local ontology feature and the at least one target local interaction feature, the original fused ontology feature under the ontology attribute dimension and the original fused interaction feature under the interaction attribute dimension are obtained through the corresponding primary feature fusion unit. Based on the original fused ontology features and the original fused interaction features, the primary fused ontology features under the ontology attribute dimension and the primary fused interaction features under the interaction attribute dimension are obtained through the corresponding primary nonlinear mapping unit.

8. The method according to claim 6, characterized in that, The advanced feature fusion submodule includes an advanced feature fusion unit and an advanced nonlinear mapping unit; The step of obtaining target fused ontology features under the ontology attribute dimension and target fused interaction features under the interaction attribute dimension based on the at least one target local ontology feature and the at least one target local interaction feature through the corresponding high-level feature fusion submodule includes: Based on the at least one target local ontology feature and the at least one target local interaction feature, the initial fused ontology feature under the ontology attribute dimension and the initial fused interaction feature under the interaction attribute dimension are obtained through the corresponding high-level feature fusion unit. Based on the initial fused ontology features and the initial fused interaction features, the target fused ontology features under the ontology attribute dimension and the target fused interaction features under the interaction attribute dimension are obtained through the corresponding high-level nonlinear mapping unit.

9. The method according to claim 1, characterized in that, The step of fusing the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object includes: The target ontology feature vectors are fused to obtain target ontology features, and the target interaction feature vectors are fused to obtain target interaction features; Based on the set fusion weights, the target local features and the target interaction features are fused to obtain the target fused features; Based on the target fusion features and the established nonlinear mapping relationship, the target value assessment information is determined.

10. The method according to claim 1, characterized in that, The method further includes: Based on the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension, the object feature description information of the target object is determined, and based on the object feature description information, operational suggestions for the target object are generated.

11. An object evaluation device, characterized in that, include: The acquisition module is configured to acquire object parameters of the target object to be evaluated in the ontology attribute dimension and the interaction attribute dimension. The feature determination module is configured to obtain the target ontology feature vector under the ontology attribute dimension and the target interaction feature vector under the interaction attribute dimension based on the object parameters under the ontology attribute dimension and the interaction attribute dimension through the corresponding feature analysis layer in the object evaluation model, wherein the feature analysis layer is determined based on the attributes of the object parameters under the corresponding attribute dimension. The object value determination module is configured to fuse the target ontology feature vector and the target interaction feature vector through the fusion layer in the object evaluation model to obtain the target value evaluation information of the target object.

12. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-10.

14. A computer program product, characterized in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-10.

Images