Dynamic heat assessment method and system based on indexes of electric smelting furnace

Abstract

The invention discloses a dynamic popularity evaluation method and system based on a danning furnace index, and the method and system enable the contribution of a user behavior to a popularity value to be exponentially attenuated as the occurrence time goes away by introducing a time attenuation factor and performing second-level processing on a behavior timestamp. According to the dynamic popularity evaluation method, the more new behaviors contribute to the final popularity value, the popularity value of a new content which rises rapidly rises can rapidly rise, the popularity value of an outdated old content can rapidly fall back, and therefore the dynamic popularity evaluation method can truly and sensitively reflect the real-time popularity trend of the content.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet information processing, and in particular to a dynamic heat evaluation method and system based on an alchemical furnace index. BACKGROUND

[0002] In today's rapid development of Internet technology, how to accurately and real-time evaluate the heat of content (such as goods, videos, news, topics) is crucial for platform content recommendation, advertisement placement, trend prediction and business decision. The existing heat evaluation method usually performs weighted summation based on the statistical quantity of user behavior data (such as clicks, views, likes, purchases). Usually, the total quantity of user behavior in a fixed time window (such as 24 hours) is calculated. This method regards the behavior data at all time points in the window as equally important, and cannot distinguish the value difference between a behavior occurring at the beginning of the window (such as 23 hours and 59 minutes ago) and a behavior occurring just now (such as 1 minute ago). As a result, a content that was once popular but whose heat has rapidly declined will continue to maintain a high ranking in the window period due to its large historical data total quantity; while a newly emerging content that is in the rapid heat rising period cannot be identified in time due to insufficient data accumulation. This leads to the fact that the heat evaluation index cannot truly and sensitively reflect the real-time popular trend of the content. SUMMARY

[0003] The present application provides a dynamic heat evaluation method and system based on an alchemical furnace index to solve the problem that the evaluation result cannot truly and sensitively reflect the real-time popular trend of the content due to the inability to distinguish the time of new and old behavior data in the existing heat evaluation method mentioned in the background.

[0004] To solve the above technical problems, in a first aspect, the present application provides a dynamic heat evaluation method based on an alchemical furnace index, characterized in that it comprises:

[0005] Data collection, collecting the behavior data publication time of users on target content, the behavior data including behavior type, user attribute and behavior scene; the publication time is an accurate second-level behavior timestamp; the behavior scene includes device type and time period characteristics;

[0006] Time decay calculation, calculating a time decay factor based on the time interval between the publication time and the current system time;

[0007] Weight distribution, constructing a dynamic weight distribution model to real-time adjust the weight coefficient of each user's behavior data;

[0008] Behavior contribution total value calculation, multiplying each user's behavior data by the corresponding weight coefficient and then accumulating to obtain the behavior contribution total value;

[0009] The heat value is calculated by multiplying the total behavior contribution value with a time decay factor.

[0010] In one embodiment, the time decay factor λ is calculated according to the formula: λ = e^(-k·Δt), where k is a decay coefficient and Δt is the time interval between the publishing time and the current system time.

[0011] In one embodiment, the method for constructing a dynamic weight distribution model and adjusting the weight coefficient of the behavior data of each user in real time comprises the following steps:

[0012] The contribution degrees of the behavior types to the content value are sorted and a basic weight is assigned;

[0013] The users are divided into three levels of high activity, medium activity and low activity according to the user attributes, and a modified behavior weight is obtained.

[0014] A scene feature coefficient is obtained in combination with the device type and the time period characteristics.

[0015] The weight coefficient is obtained by multiplying the basic weight, the modified behavior weight and the scene feature coefficient.

[0016] In one embodiment, when the cumulative number of new behavior data is greater than a preset threshold or the time interval is greater than a preset time, the system automatically recalculates the time decay factor and determines whether the weight coefficient needs to be dynamically adjusted.

[0017] In one embodiment, the method for determining whether the weight coefficient needs to be dynamically adjusted comprises the following steps:

[0018] When the proportion of a certain type of user behavior in a single content exceeds a preset threshold, or the user activity level changes, or the device type or time period of the user behavior changes, the weight coefficient is recalculated, and the adjustment result is applied to the calculation of the heat value.

[0019] In a second aspect, the application further provides a dynamic heat evaluation system based on the alchemical furnace index, which is characterized by being applied to the dynamic heat evaluation method based on the alchemical furnace index.

[0020] A data acquisition module is configured to acquire behavior data and publishing time of a user on target content, and the data acquisition module, as a system data inlet, synchronously outputs the acquired behavior data and publishing time to a time decay calculation module and a user behavior weight distribution module to provide basic data support for subsequent calculation.

[0021] a time decay calculation module configured to receive a content publishing time output by the data acquisition module, and calculate a time decay factor based on a time interval between the content publishing time and a current system time; the decay factor decreases dynamically with an increase in the time interval, ensuring that the content popularity evaluation conforms to the timeliness rule over time;

[0022] a user behavior weight distribution module configured to receive behavior type, user attribute, and behavior scene data output by the data acquisition module, and dynamically distribute weight coefficients of the behavior type, user attribute, and behavior scene;

[0023] a popularity value calculation module configured to receive the decay factor output by the time decay calculation module and the weight coefficients output by the user behavior weight distribution module, and calculate a popularity value of the target content.

[0024] In one of the embodiments, the user behavior weight distribution module further includes a behavior quality evaluation unit configured to perform semantic analysis on the comment behavior, and adjust weight values of positive sentiment comments and negative sentiment comments.

[0025] In one of the embodiments, the evaluation system further includes a dynamic update module configured to monitor and receive new user behavior data and time intervals in real time, and trigger the system to enter a dynamic update process:

[0026] The dynamic update module includes a behavior data judgment unit, a time interval processing unit, and a weight coefficient processing unit.

[0027] The behavior data judgment unit is configured to judge whether the cumulative number of new behavior data is greater than a preset threshold, and when the cumulative number of new behavior data is greater than the preset threshold, trigger the system to enter the dynamic update process;

[0028] The time interval processing unit is configured to, when the time interval is greater than a preset time, trigger the system to enter the dynamic update process and automatically recalculate the time decay factor;

[0029] The weight coefficient processing unit is configured to, when the proportion of a certain type of user behavior in a single content exceeds a preset threshold, or the user activity level changes, or the user behavior switches the device type or the time period, recalculate the weight coefficient.

[0030] In one of the embodiments, the dynamic update module further includes an adaptive computing power scheduling unit, which includes a popularity level division subunit, a dynamic switching subunit, and a hierarchical scheduling strategy subunit.

[0031] The popularity level division subunit is configured to divide the popularity value into three levels of high popularity, medium popularity, and low popularity.

[0032] The dynamic switching subunit is configured to monitor content heat level changes in real time, and trigger resource switching when the heat level crosses a threshold value.

[0033] The hierarchical scheduling strategy subunit is configured to preferentially allocate GPU core modules to high-heat content, CPU core modules to medium-heat content, and adopt a 30-second interval batch processing mode for low-heat content.

[0034] In one of the embodiments, the dynamic updating module further comprises an abnormal behavior detection unit configured to filter repeated behaviors generated by the same IP within a preset time period and greater than a preset number of times, so as to reduce the influence of data noise.

[0035] In a third aspect, the present application further provides a computer device comprising a processor and a memory, wherein the memory is configured to store a computer program, and the computer program is configured to be executed by the processor to implement the dynamic heat evaluation method and system based on the alchemical furnace index.

[0036] In a fourth aspect, the present application further provides a computer readable storage medium storing a computer program, and the computer program is configured to be executed by a processor to implement the dynamic heat evaluation method and system based on the alchemical furnace index.

[0037] Compared with the prior art, the present application has at least the following beneficial effects:

[0038] The dynamic heat evaluation method and system based on the alchemical furnace index can truly and sensitively reflect the real-time popular trend of content. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 FIG. 1 is a flowchart of the dynamic heat evaluation method based on the alchemical furnace index according to an embodiment of the present application;

[0040] Figure 2 FIG. 2 is a structural diagram of the dynamic heat evaluation system based on the alchemical furnace index according to an embodiment of the present application;

[0041] Figure 3 FIG. 3 is a structural diagram of the computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0042] With reference to the drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the present application of a dynamic heat evaluation method and system based on an alchemical furnace index is only for the purpose of describing specific embodiments, and is not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0044] In the digital age of information explosion, the scientific nature of content heat evaluation mechanism directly affects the accuracy of information distribution and user experience. The static evaluation mechanism commonly used in the prior art mainly relies on cumulative statistics within a fixed time window (such as the total number of clicks or interactions in the past 24 hours or 7 days) and a design logic that does not automatically decay with time. This mechanism has significant defects in timeliness and is difficult to adapt to the objective law of dynamic changes in content value.

[0045] The core problem of the static evaluation mechanism is that its time dimension processing method is out of line with the inherent characteristics of the content life cycle. The information value of content usually follows a certain decay law - most content has a high degree of attention and information density in the early stage of publication, and its timeliness value will show an exponential or linear decay trend (such as the exponential decay of breaking news and the linear decay of technical documents). However, the static evaluation mechanism treats the statistical data within the time window as a fixed weight, and cannot distinguish the actual value difference of content at different publication times within the same time window. For example, in the 7-day fixed window statistical mode, high-interactive content published 3 days ago and similar content published 1 hour ago may be given the same heat weight, resulting in old content occupying the hot spot position due to the cumulative data advantage, while new content with real-time value is submerged.

[0046] This mechanism defect directly leads to two levels of negative effects: on the one hand, the "hot information" obtained by the user may have lost its timeliness value, forming an "information lag" phenomenon, especially in scenarios such as news information, social media, real-time collaboration, which have very high requirements for timeliness, users may make wrong judgments or miss critical opportunities due to outdated information; on the other hand, the dynamic balance of the platform content ecosystem is destroyed, and the long-term occupation of the hot spot position by old content will squeeze the exposure space of new content, reducing the metabolic efficiency of the information ecosystem, and ultimately leading to a decline in user trust in platform information.

[0047] For example, old content decay: an article published 3 days (Δt = 72 hours) (k = 0.02 / hour) of information, λ = e^(-0.02x72) ≈ 0.24, even if S = 1000 (high interaction), H = 1000x0.24 = 250;

[0048] New content exposure: a breaking news published 10 minutes (Δt = 0.17 hours) (k = 0.02 / hour), λ ≈ e^(-0.02x0.17) ≈ 0.997, even if S = 300 (low interaction), H = 300x0.997 ≈ 299, the heat value is higher than the old content.

[0049] In summary, the dynamic heat evaluation method of the present application can truly and sensitively reflect the real-time popular trend of the content.

[0050] The alchemical furnace index is a dynamic updated quantitative evaluation index based on multidimensional big data and artificial intelligence algorithm, which is used to measure the attention degree and popular trend of specific target objects (such as goods, content, brand, topic, etc.) in the network environment in real time. Through a set of comprehensive calculation method, a single score reflecting real-time heat and value quality is obtained from a large amount of chaotic user behavior data.

[0051] The alchemical furnace index mainly focuses on the sales and sales of goods, and purely carries out sales data analysis, sales situation analysis and equipment trend analysis, without exposing the customer information of each platform.

[0052] Please refer to Figure 1 The present application provides a dynamic heat evaluation method based on the alchemical furnace index, comprising

[0053] S1, data acquisition, acquiring behavior data b_i and publishing time of users to target content, the behavior data including behavior type, user attribute and behavior scene; the publishing time is accurate to second level behavior timestamp; the behavior scene includes device type and time period characteristics;

[0054] S2, time decay calculation, based on the time interval Δt between the publishing time and the current system time, calculate the time decay factor λ; the publishing time is set as t1, the current system time is t2, then the time interval Δt = t2-t1; the calculation formula of the time decay factor λ is: λ = e^(-k·Δt), wherein k is the decay coefficient, which can be dynamically adjusted according to the content type, such as k value of information content is 0.015-0.03 / hour, which is higher than 0.005-0.015 / hour of tool content, Δt: the time interval between publishing time and current system time.

[0055] S3, weight allocation, constructing a dynamic weight allocation model, and adjusting the weight coefficient w_i of the behavior data of each user in real time;

[0056] The method for constructing a dynamic weight allocation model and adjusting the weight coefficient w_i of the behavior data of each user in real time is:

[0057] S31, sort the contribution of the content value according to the behavior type, and give a basic weight;

[0058] For example, the basic weight of purchase is 10, the basic weight of adding is 6, the basic weight of sharing is 5, the basic weight of collecting is 4, the basic weight of liking is 2, and the basic weight of browsing is 1;

[0059] S32, divide the users into high, medium and low active levels according to the user attributes, and obtain the modified behavior weight;

[0060] For example, core users and VIPs are high active users, the correction coefficient = 1.5; ordinary users are medium active users, the correction coefficient = 1.0; new users or silent users are low active users, the correction coefficient = 0.8;

[0061] S33, combine the device type and time period characteristics to obtain the scene characteristic coefficient;

[0062] Different devices and different time periods may reflect different input and intentions of the user. For example, the device type is a mobile phone, the user may use fragmented time, and the attention is easy to scatter, so the device type coefficient = 1.0; the device type is a computer, the user may be more focused and the intention may be more clear, so the device type coefficient = 1.2;

[0063] Regarding the time period characteristics, if the user operates in the golden time period (20:00-22:00), it is leisure time, the intention is real, and the time period characteristic coefficient = 1.2; if the user operates in the late night time period (00:00-06:00), the user may be in the period of irrational consumption, or just randomly browse, and the time period characteristic coefficient = 0.9; if the user operates in the working time period (09:00-18:00), the user may steal time from work, and the decision is hasty, and the time period characteristic coefficient = 1.0;

[0064] Therefore, the scene characteristic coefficient = device type coefficient x time period characteristic coefficient. For example, user A browses with his PC computer at 21:30 on Wednesday night. The device type is a computer, the device type coefficient = 1.2, and the time period characteristic 21:30 is the golden time period, the time period characteristic coefficient = 1.2, so the scene characteristic coefficient = 1.2 x 1.2 = 1.44.

[0065] S34, obtain the weight coefficient w_i = basic weight x modified behavior weight x scene characteristic coefficient.

[0066] For example, if user A is a high active user, the correction factor = 1.5, the basic weight of collection is 4, and the scene feature coefficient is 1.44, so the weight coefficient w_i = 4 x 1.5 x 1.44 = 8.64.

[0067] S4, behavior contribution total value calculation, multiply each user behavior data b_i with the corresponding weight coefficient w_i and accumulate to obtain the behavior contribution total value S = Σ(b_i·w_i);

[0068] b_i: represents the i-th user behavior data. In actual calculation, it is usually a count. For example, one purchase behavior, b_i = 1; one like behavior, b_i = 1. Its core function is to record that the behavior has occurred once. w_i: represents the weight coefficient corresponding to the i-th behavior.

[0069] For example, we evaluate the scene of a camping short video heat. Suppose the system has collected the following 6 user behaviors in the past 1 hour, as shown in the following table:

[0070] We add up the contribution values of the 6 behaviors to obtain the behavior contribution total value S, S = 8.64 + 2.88 + 15.00 + 3.00 + 7.20 + 0.50 = 37.22.

[0071] The statistical method in the prior art only tells you that the user has performed 6 behaviors, while the model of the present application tells you that the total value of the 6 behaviors is 37.22. One high-quality purchase behavior (value 15) has a much greater impact than multiple low-quality “browsing” behaviors (value 0.5). This makes the final heat value better reflect the real user intention and commercial value.

[0072] S5, heat value calculation, multiply the behavior contribution total value by the time decay factor to obtain the heat value H = S x λ.

[0073] The behavior contribution total value S represents the highest heat value that the content can achieve under ideal conditions, and the time decay factor λ decreases over time, so H = S x λ represents the real-time heat of the content, which is equal to its behavior contribution total value S, and the remaining part after time decay is the current heat value.

[0074] In one embodiment, when the number of newly added behavior data is greater than a preset threshold or the time interval is greater than a preset time, the system automatically recalculates the time decay factor λ and determines whether the weight coefficient w_i needs to be dynamically adjusted.

[0075] The system sets a counter (e.g., N=1000). Each time a new user behavior (e.g., likes, purchases) occurs, the counter is incremented by 1. When the cumulative number of new behaviors exceeds 1000, the time decay factor λ is immediately recalculated. This ensures that when the content popularity changes significantly, the system can perceive and update the results in a timely manner.

[0076] The system also has a time trigger (e.g., T=5 minutes). Regardless of the data volume change, if it has been more than 5 minutes since the last calculation, a recalculation of the time decay factor λ is triggered. Some content may have slow data growth and may not reach the quantity threshold for a long time, but its popularity is also changing slowly. This time trigger is equivalent to providing a bottom-line mechanism to ensure that the popularity values of all content are updated regularly.

[0077] Once either of the above conditions is triggered, the system performs the following two actions: First, the system obtains the current latest time, recalculates the time difference Δt, and substitutes it into the calculation formula of the time decay factor λ: λ=e^(-k*Δt) to generate an updated decay factor based on the current time for each piece of data. Second, it determines whether the weight coefficient w_i needs to be dynamically adjusted.

[0078] The method for determining whether the weight coefficient w_i needs to be dynamically adjusted is:

[0079] When the proportion of a certain type of user behavior in a single piece of content exceeds the preset threshold, or the user activity level changes, or the user behavior switches devices or time periods, the weight coefficient w_i is recalculated, and the adjustment result is applied to the calculation of the popularity value H.

[0080] Suppose we have been evaluating the popularity of a newly released smartwatch. For example, if the purchase behavior accounts for more than 20% of the new data, the adjustment is triggered. When the proportion of purchase behavior in the new data increases from 5% to 35%, the system may increase the weight of purchase behavior, for example, from 10 to 12, and the weight coefficient w_i is adjusted accordingly.

[0081] When the distribution of user activity levels changes, for example, the proportion of high-active users in new users jumps from 10% to 60%, the system will assign a higher correction coefficient to all behaviors of high-active users, for example, from 1.5 to 1.8, and the weight coefficient w_i is adjusted accordingly.

[0082] When the user behavior occurs device type switching, for example, the device proportion from PC end is greatly improved from 20% to 65%, the system will improve the device type coefficient of PC end behavior, for example, from 1.2 to 1.5, the scene feature coefficient will also change, and then the weight coefficient w_i will also be adjusted together.

[0083] When the time period distribution of user behavior changes, for example, in the new behavior, the proportion of occurrence in the night golden time period is increased from 30% to 70%, the system will improve the time period feature coefficient of the golden time period, for example, from 1.2 to 1.4, the scene feature coefficient will also change, and then the weight coefficient w_i will also be adjusted together.

[0084] Please refer to Figure 2 , in the second aspect, the application also provides a dynamic heat evaluation system based on the alchemical furnace index, which is applied to the dynamic heat evaluation method based on the alchemical furnace index, and the system comprises:

[0085] The data acquisition module 1 is used for acquiring the behavior data and the publishing time of the user to the target content, the data acquisition module is used as a system data inlet, the acquired behavior data and the publishing time are synchronously output to the time decay calculation module and the user behavior weight distribution module, and the basic data support is provided for subsequent calculation;

[0086] The acquisition mode is: real-time acquisition through front-end burying points (APP / webpage SDK), server logs (Nginx / Apache), database change monitoring (MySQLBinlog) multiple channels; the accurate time stamp of content creation is accurate to seconds, such as "2025-08-2310:30:05";

[0087] Data preprocessing: removing duplicate data (such as 5 times of continuous clicks of the same user within 1 second), and marking abnormal data, for example, simulator behavior, to ensure the quality of input data;

[0088] Kafka message queue is adopted to realize low-delay transmission (delay≤100ms), and is output to the time decay calculation module 2 and the user behavior weight distribution module 3;

[0089] The time decay calculation module 2 is used for receiving the content publishing time output by the data acquisition module, calculating the time decay factor based on the time interval between the content publishing time and the current system time; the decay factor decreases dynamically with the increase of the time interval, to ensure that the content heat evaluation conforms to the timeliness law with the passage of time;

[0090] The publishing time is set as t1, the current system time is t2, and the time interval At = t2-t1; the calculation formula of the time decay factor λ is: λ = e^(-k At), wherein k is the decay coefficient, and At is the time interval between the publishing time and the current system time.

[0091] The user behavior weight distribution module 3 is configured to receive the behavior type, user attribute, and behavior scene data output by the data collection module, and dynamically distribute the weight coefficients of the behavior type, user attribute, and behavior scene.

[0092] The heat value calculation module 4 is configured to receive the decay factor output by the time decay calculation module and the weight coefficient output by the user behavior weight distribution module, and obtain the heat value of the target content through calculation.

[0093] The time decay factor λ and the weight coefficient w_i are integrated, and the final heat value H is generated through calculation, which is the decision output unit of the system.

[0094] In one embodiment, the user behavior weight distribution module 3 further includes a behavior quality evaluation unit configured to perform semantic analysis on the comment behavior and adjust the weight values of positive and negative sentiment comments.

[0095] The behavior quality evaluation unit is based on a natural language processing model and is configured to understand the meaning and sentiment of the comment text. The system dynamically assigns different weight values to comments with different sentiments according to the results of semantic analysis.

[0096] Suppose two users have commented on the same smart phone. In the basic model, the basic weight of the two comment behaviors may be 3. If the comment content of user A is "This phone is really great! The camera takes clear photos, the battery lasts a long time, and there is still 30% of the battery left after a day of use. Strongly recommended!", the behavior quality evaluation unit judges it as a positive sentiment comment. According to the preset rules of the behavior quality evaluation unit, the contribution of positive sentiment comments to heat is positive, and a higher weight should be assigned. Therefore, the final basic weight value of this comment behavior is no longer 3, but is adjusted to 3*1.5=4.5, assuming that the positive coefficient is 1.5.

[0097] In one embodiment, the evaluation system further includes a dynamic update module configured to monitor and receive new user behavior data and time intervals in real time, and trigger the system to enter a dynamic update process.

[0098] The dynamic update module includes a behavior data judgment unit, a time interval processing unit, and a weight coefficient processing unit.

[0099] The behavior data judging unit is configured to judge whether the cumulative number of the newly added behavior data is greater than a preset threshold value, and trigger the system to enter the dynamic updating process when the cumulative number of the newly added behavior data is greater than the preset threshold value.

[0100] The time interval processing unit is configured to trigger the system to enter the dynamic updating process and automatically recalculate the time decay factor λ when the time interval is greater than a preset time.

[0101] The weight coefficient processing unit is configured to recalculate the weight coefficient w_i when the proportion of a certain type of user behavior in a single piece of content exceeds a preset threshold value, or the user activity level changes, or the user behavior changes the device type, or the time period changes.

[0102] In one of the embodiments, the dynamic updating module further comprises an adaptive computing power scheduling unit, which comprises a heat level division subunit, a dynamic switching subunit, and a hierarchical scheduling strategy subunit.

[0103] The heat level division subunit is configured to divide the heat value H into three levels of high heat, medium heat, and low heat.

[0104] For example, we define the heat value H<1000 as low heat, 1000≤heat value H<10000 as medium heat, and heat value H≥10000 as high heat.

[0105] The dynamic switching subunit is configured to monitor the content heat level changes in real time, and trigger resource switching when the heat level crosses the threshold value.

[0106] The dynamic switching subunit continuously monitors the heat value H of all contents. When it is found that the heat value H of a certain content crosses the boundary from one level interval to another, for example, from 999 to 1000, or from 10050 to 9990, a resource switching instruction is triggered immediately.

[0107] The hierarchical scheduling strategy subunit is configured to preferentially allocate GPU core modules to high-heat content, allocate CPU core modules to medium-heat content, and use a 30-second interval batch processing mode for low-heat content.

[0108] The hierarchical scheduling strategy subunit receives the resource switching instruction, and executes the preset resource scheduling strategy according to the current heat level of the content. For high-heat content, GPU core modules are preferentially allocated for calculation. Because GPU has thousands of cores and is good at parallel processing of large amounts of data, high-heat content means that there are massive new behavior data flowing in every second, which needs to be calculated at the fastest speed to obtain time decay factor and weight coefficient, to ensure the real-time performance of the heat value H. For medium-heat content, CPU core modules are allocated for calculation, because the number of CPU cores is small, but they are good at complex logic and serial processing. The data of medium-heat content grows steadily, and does not require extreme real-time performance. Using CPU core modules to process can meet the demand, and the cost is much lower than that of GPU core modules. For low-heat content, a 30-second interval batch processing mode is adopted. The heat of these contents changes slowly, and almost no one pays attention to them. It is a huge waste to allocate real-time computing resources to them. The system will package their data, and calculate them once every 30 seconds.

[0109] In one of the embodiments, the dynamic updating module further comprises an abnormal behavior detection unit, which filters repeated behaviors generated by the same IP more than a preset number of times within a preset time, to reduce the influence of data noise.

[0110] The abnormal behavior detection unit monitors all the inflowing behavior data in real time, and extracts the IP address. It also counts how many times the same type of behavior, such as likes and comments, is generated by the same IP address within a recent preset time T. When the abnormal behavior detection unit finds that the number of repeated behaviors of the same IP within time T is greater than the preset number N, it is determined that the behavior of the IP is most likely abnormal behavior. The abnormal behavior detection unit will discard or greatly reduce the weight of these repeated behaviors, so that they do not enter or only enter the subsequent weight and heat calculation process at a very low rate.

[0111] For example, evaluate the heat of a new game. The system has preset rules: if the same IP generates more than 30 likes within 1 minute, it is considered abnormal. A game guild of 20 real players organizes an activity in an internet cafe. Since the internet cafe uses the same public IP, the 20 people have all liked the game within 1 minute. The abnormal behavior detection unit detects that the like behavior from a certain IP address has reached 20 times within 1 minute. However, 20 times is less than the preset threshold of 30 times, so it is determined to be normal behavior, and all 20 like behaviors participate in heat calculation normally.

[0112] The dynamic heat evaluation system based on the alchemical furnace index further comprises a reputation value calculation module for correcting the heat value and obtaining a new heat value formula.

[0113] H 新=S x l x (1 + V), l: time decay factor, S: total value of behavior contribution, V: net value of reputation. The better the reputation (V is positive and large), the higher the final heat; the worse the reputation (V is negative), the final heat will be inhibited. ;∑: summation symbol, representing the accumulation of all comments within the preset time window; S i : sentiment score of the i-th comment. This is a quantitative sentiment value calculated by an NLP model, usually ranging from -1.0 (extremely negative) to +1.0 (extremely positive).W j =W base x K quality ;W base is the base weight of the comment behavior (e.g., a fixed value of 3), K quality is the quality coefficient, for example, comment length: longer comments usually contain more information, and the quality coefficient is higher (e.g., comment > 50 words, coefficient = 1.2).W j x S i , i.e., the contribution value of a single comment. The contribution values of all comments within the time window are accumulated, i.e., the final net value of reputation V is obtained.

[0114] Suppose we calculate the net value of reputation of a Bluetooth headset for the past 7 days. User A's comment: "The sound quality is very good!". Then the sentiment score S1 = +0.7, the comment is short, the quality coefficient K quality = 0.9, and if the base weight is 3, then the comprehensive weight W1 = 3 x 0.9 = 2.7, V = 2.7 x 0.7 = 1.89. The net value of reputation V serves as a quantitative bridge to integrate unstructured text comment information and structured user behavior data within the same mathematical framework (heat calculation model). This makes the system's decision-making more comprehensive and intelligent.

[0115] Please refer to Figure 3 , in a third aspect, the present application also provides a computer device 5, comprising a processor 50 and a memory 51, the memory 51 is used to store a computer program 52, the computer program 52 is executed by the processor 50 to realize the dynamic heat evaluation method and system based on the alchemical furnace index.

[0116] The computer device 5 can be a tablet computer, a desktop computer and a cloud server and the like. The computer device can include but is not limited to processor 50, memory 51. Those skilled in the art can understand that Figure 3 The computer device 5 is only an example and does not constitute a limitation on the computer device 5, which can include more or fewer components than the illustration, or combine certain components, or different components, for example, it can also include input / output devices, network access devices, etc.

[0117] The processor 50 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0118] The memory 51 can be an internal storage unit of the computer device 5 in some embodiments, such as a hard disk or a memory of the computer device 5. The memory 51 can also be an external storage device of the computer device 5 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 51 can include both an internal storage unit and an external storage device of the computer device 5. The memory 51 is used to store an operating system, an application program, a Boot Loader, data, and other programs, such as program codes of the computer program, etc. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0119] The technical features of the above embodiments can be combined in any manner. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described, but it should be understood that any combination of the technical features is within the scope of the present disclosure as long as the combination does not result in contradictions.

[0120] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these are within the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A dynamic heat assessment method based on the alchemy furnace index, characterized in that, include: Data collection involves collecting user behavior data related to the target content, including the posting time. This behavior data includes behavior type, user attributes, and behavior scenario. The publication time is a behavior timestamp accurate to the second. The behavioral scenarios include device type and time period characteristics; The time decay calculation is performed by calculating a time decay factor based on the time interval between the release time and the current system time. Weight allocation: Construct a dynamic weight allocation model to adjust the weight coefficients of each user's behavioral data in real time. The total value of behavioral contribution is calculated by multiplying each user's behavioral data by its corresponding weight coefficient and then summing the results. The heat value is calculated by multiplying the total contribution of the behavior by the time decay factor. The method for constructing a dynamic weight allocation model and adjusting the weight coefficients of each user's behavioral data in real time is as follows: The content value is ranked according to the contribution of the aforementioned behavior types, and a basic weight is assigned to each type. Users are categorized into three levels—highly active, moderately active, and lowly active—based on their attributes, and the corrected behavior weights are obtained. The scene characteristic coefficients are derived by combining equipment type and time period characteristics; The weight coefficient is calculated as: base weight × modified behavior weight × scene feature coefficient.

2. The dynamic heat assessment method based on the alchemy furnace index as described in claim 1, characterized in that, The formula for calculating the time decay factor λ is: λ=e^(-k·Δt), where k is the decay coefficient and Δt is the time interval between the release time and the current system time.

3. The dynamic heat assessment method based on the alchemy furnace index as described in claim 1, characterized in that, When the cumulative number of new behavioral data exceeds a preset threshold or the time interval exceeds a preset time, the system automatically recalculates the time decay factor and determines whether the weight coefficient needs to be dynamically adjusted.

4. The dynamic heat assessment method based on the alchemy furnace index as described in claim 3, characterized in that, The method for determining whether the weighting coefficients need to be dynamically adjusted is as follows: When the proportion of a certain type of user behavior in a single piece of content exceeds a preset threshold, or when the user's activity level changes, or when the user's behavior changes due to a device type switch or time period switch, the weight coefficient is recalculated, and the adjustment result is applied to the calculation of the popularity value.

5. A dynamic heat assessment system based on the alchemy furnace index, characterized in that, A system applied to the dynamic thermal assessment method based on the alchemy furnace index as described in any one of claims 1 to 4, the system comprising: The data acquisition module is used to collect user behavior data and publication time of target content. As the system data entry point, the data acquisition module synchronously outputs the collected behavior data and publication time to the time decay calculation module and the user behavior weight allocation module, providing basic data support for subsequent calculations. The time decay calculation module receives the content release time output by the data acquisition module and calculates the time decay factor based on the time interval between the content release time and the current system time. The decay factor decreases dynamically as the time interval increases, ensuring that the content popularity assessment conforms to the timeliness law over time. The user behavior weight allocation module is used to receive behavior type, user attribute and behavior scenario data output by the data acquisition module, and dynamically allocate weight coefficients for behavior type, user attribute and behavior scenario. The popularity value calculation module receives the decay factor output by the time decay calculation module and the weight coefficient output by the user behavior weight allocation module, and calculates the popularity value of the target content.

6. The dynamic thermal evaluation system based on the alchemy furnace index as described in claim 5, characterized in that, The user behavior weight allocation module also includes a behavior quality assessment unit, which performs semantic analysis on comment behavior and adjusts the weight values ​​of positive and negative sentiment comments.

7. The dynamic thermal evaluation system based on the alchemy furnace index as described in claim 5, characterized in that, The evaluation system also includes a dynamic update module, which is used to monitor and receive new user behavior data and time intervals in real time, triggering the system to enter the dynamic update process: The dynamic update module includes a behavior data judgment unit, a time interval processing unit, and a weight coefficient processing unit. The behavior data judgment unit is used to determine whether the cumulative number of newly added behavior data is greater than a preset threshold. When the cumulative number of newly added behavior data is greater than the preset threshold, the system is triggered to enter the dynamic update process. The time interval processing unit is used to trigger the system to enter the dynamic update process and automatically recalculate the time decay factor when the time interval is greater than the preset time. The weighting coefficient processing unit is used to recalculate the weighting coefficient when the proportion of a certain type of user behavior in a single piece of content exceeds a preset threshold, or when the user's activity level changes, or when the user's behavior changes device type or time period.

8. The dynamic thermal assessment system based on the alchemy furnace index as described in claim 7, characterized in that, The dynamic update module also includes an adaptive computing power scheduling unit, which includes a heat level division subunit, a dynamic switching subunit, and a hierarchical scheduling strategy subunit. The heat level division subunit is used to divide the heat value into three levels: high heat, medium heat, and low heat. The dynamic switching subunit is used to monitor changes in content popularity level in real time, and trigger resource switching when the popularity level crosses the threshold. The hierarchical scheduling strategy subunit is used to prioritize the allocation of high-popularity content to GPU core modules, medium-popularity content to CPU core modules, and low-popularity content to a batch processing mode with a 30-second interval.

9. The dynamic heat assessment system based on the alchemy furnace index as described in claim 7, characterized in that, The dynamic update module also includes an abnormal behavior detection unit, which filters out repeated behaviors generated by the same IP more than a preset number of times within a preset time to reduce the impact of data noise.

Images