User Portrait Tag Evaluation Method

Abstract

The present invention discloses a method for evaluating user portrait tags, including: constructing an intelligent creative component according to the tags to be evaluated; using an automated bottoming engine to issue the intelligent creative component, completing the issuance of the portrait posterior component of the target group, and obtaining an experimental group and a control group; performing data mining on the click behavior of the target group to calculate the actual number of clicking users A; quantifying the indicators of the experimental group and the control group to obtain the accuracy and importance of the tags; evaluating the portrait tags according to the accuracy and importance of the tags, and inferring whether the tag is important for the corresponding business, that is, whether it is necessary to continuously maintain the tag; the present invention not only evaluates according to the click-through rate and accuracy rate, but also defines the importance of the tags, etc., making the tag evaluation more comprehensive and accurate.

Description

Technical Field

[0001] The present invention relates to the technical fields of big data and artificial intelligence, and in particular to a method for evaluating user portrait tags. Background Art

[0002] A tag is a symbolic representation of a certain user characteristic. It is a content organization method and a keyword with strong relevance, which can conveniently help us find appropriate content and content classification. Scientifically and comprehensively evaluating the quality of tags helps to control the tag quality and guides tag managers and developers to continuously improve the tag quality. By creating a complete evaluation system, for tags with extremely poor quality, it can be considered not to be launched, and it can only be opened for business use after meeting the basic quality requirements. Otherwise, it will neither bring value to the business nor easily make the tag portrait system lose users' trust.

[0003] Currently, the methods for evaluating user portrait tags mainly rely on existing user research feedback data or posterior click-throughs of applications to check whether the click-through rate and accuracy rate reach the target. However, analyzing only based on the accuracy rate cannot well reflect the importance and quality of tags. Summary of the Invention

[0004] To solve the problems existing in the prior art, the purpose of the present invention is to provide a method for evaluating user portrait tags. The present invention not only evaluates based on the click-through rate and accuracy rate, but also defines the importance of tags, etc., making the tag evaluation more comprehensive and accurate.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is: a method for evaluating user portrait tags, comprising the following steps:

[0006] Step 1: Construct an intelligent creative component according to the tag to be evaluated;

[0007] Step 2: Use an automated bottoming engine to distribute the intelligent creative component, complete the distribution of the portrait posterior component of the target group, and obtain an experimental group and a control group;

[0008] Step 3: Conduct data mining on the click behavior of the target group and calculate the actual number of clicking users A;

[0009] Step 4: Quantify the indicators of the experimental group and the control group to obtain the accuracy and importance of the tag;

[0010] Step 5: Evaluate the portrait tag according to the accuracy and importance of the tag, and infer whether the tag is important for the corresponding business, that is, whether it is necessary to continuously maintain the tag.

[0011] As a further improvement of the present invention, the specific content of Step 1 includes:

[0012] Infer relevant tags of the tag based on user behavior data and requirements;

[0013] Develop relevant intelligent creative components according to the relevant tags to attract users.

[0014] As a further improvement of the present invention, step 2 specifically includes:

[0015] Send the intelligent creative components to each intelligent terminal for users to click according to requirements, and collect relevant click data;

[0016] For the inferred relevant tags, perform personalized user screening respectively, place the inferred relevant tags at the top, and randomly distribute other tags in order, and set it as the experimental group;

[0017] Screen a part of users from the remaining users and randomly send relevant components, and set it as the control group.

[0018] As a further improvement of the present invention, the number of users screened in the experimental group and the control group is the same, and is set as M.

[0019] As a further improvement of the present invention, in step 3, the actual number of clicking users A is the sum of the de-duplicated number of clicking users for each component.

[0020] As a further improvement of the present invention, in step 4, the quantification of the indicators of the experimental group and the control group specifically includes:

[0021] Calculate the click accuracy rate P: P = actual number of clicking users / (actual number of clicking users + number of mis-inferred clicking users) = actual number of clicking users / number of sent users = A / M;

[0022] Calculate the click-through rate PR of the inferred relevant tags among different groups of users: PR = actual number of clicking users / number of sent users = A / M;

[0023] Calculate the gain of the number of tag users: gain of the number of tag users = PR (experimental group) - PR (control group) = A (experimental group) / M - A (control group) / M = (A (experimental group) - A (control group)) / M;

[0024] Calculate the tag importance S: S = gain of the number of tag users / accuracy rate = (A (experimental group) - A (control group)) / (M * (A / M)) = (A (experimental group) - A (control group)) / A;

[0025] Since A (experimental group) is the number of users who actually click on the inferred tag, which is the same as the actual number of clicking users A, the tag importance S can be simplified as: S = 1 - A (control group) / A. If A (control group) / A is higher, it means that this tag is less important, that is, even without inferring this tag, users will still click.

[0026] The beneficial effects of the present invention are as follows:

[0027] The present invention calculates the label gain according to the experimental group and the control group, and further defines the importance of the label in combination with the accuracy rate, etc., making the label evaluation more comprehensive and accurate, effectively evaluating the quality of the inferred label, and then optimizing each label algorithm. Brief Description of the Drawings

[0028] Figure 1 is a flowchart of an embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of the division structure of the experimental group and the control group in an embodiment of the present invention. Detailed Description of the Embodiment

[0030] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0031] Embodiment 1

[0032] As Figure 1 shown, a method for evaluating user portrait labels includes:

[0033] (1) Construct an intelligent creative component for the label to be evaluated:

[0034] Infer the relevant labels of the label according to the user behavior data and requirements;

[0035] Develop relevant intelligent creative components according to the relevant labels to attract users.

[0036] (2) Use an automated bottoming engine to complete the distribution of the portrait posterior component of the target group:

[0037] Distribute the intelligent creative components to each TV terminal according to the requirements for users to click, and collect relevant click data;

[0038] As Figure 2 shown, for the inferred relevant labels, perform personalized user screening respectively, place the inferred relevant labels at the top, and randomly distribute the other labels in order, which is set as the experimental group;

[0039] Select a part of users from the remaining users and randomly distribute the relevant components, which is set as the control group;

[0040] The number of users screened in the experimental group is the same as that in the control group, which is set as M.

[0041] (3) Conduct data mining and analysis on the click behavior of the target group:

[0042] Calculate the actual number of clicking users: The actual number of clicking users (A) is the sum of the number of clicking users of each component after removing duplicates.

[0043] (4) Quantification of indicators for the control group and the experimental group:

[0044] Accuracy rate (P) = Number of actual click users / (Number of actual click users + Number of misjudged click users) = Number of actual click users / Number of sent users = A / M;

[0045] Calculate the click-through rate PR of the number of users for each label in different groups, where the label is several inference labels in step (1), PR = Number of actual click users / Number of sent users = A / M;

[0046] Gain of label user number = PR (experimental group) - PR (control group) = A (experimental group) / M - A (control group) / M = (A (experimental group) - A (control group)) / M;

[0047] Output the importance of the label, more precisely:

[0048] Importance of label S = Label gain / Accuracy rate = (A (experimental group) - A (control group)) / (M * (A / M)) = (A (experimental group) - A (control group)) / A;

[0049] Then A (experimental group) is the number of users who actually click on the inference label, which is consistent with the number of actual click users A. Therefore, the above formula S = 1 - A (control group) / A. If A (control group) / A is higher, it means that this label is less important, that is, even without this inference label, users will still click.

[0050] (5) Conduct portrait label evaluation:

[0051] According to the importance and accuracy of the label, infer whether the label is important for this business, that is, whether it is necessary to continuously maintain the label, so as to reduce the maintenance workload and use more energy to maintain the labels with strong importance, which is of great significance for the subsequent continuous update of the labels.

[0052] Example 2

[0053] As Figure 1 shown, a method for evaluating user portrait labels includes:

[0054] (1) Construct an intelligent creative component portrait library for the label to be evaluated:

[0055] Based on the viewing behavior data of users, infer whether there are elderly people, middle-aged men, middle-aged women, adult men, adult women, children, etc. in the user's family, and construct creative components. The component names are: Retired people, Real men, Curiosity, Girlish heart, Moms, Cute kids. The constructed intelligent creative component portrait library is shown in the following table:

[0056] Inference label Elderly Middle-aged man Middle-aged woman Adult man Adult woman Child Creative component Retirees Real man Stay-at-home mom Curiosity Girlish heart Cute baby

[0057] (2) After completing the portrait posterior component distribution of the target group using the automated baseline engine, as Figure 2 shown, set up an experimental group and a control group:

[0058] Distribute the intelligent creative components to each TV terminal according to requirements for users to click, and collect relevant click data for analysis;

[0059] Select a part of the users for experiments based on the inferred tags. For example, if it is inferred that there are elderly people among this group of users, then the first order of the components for the user group containing the elderly is the elderly component, and the other tags are randomly distributed, set as the experimental group;

[0060] Select a part of the users from the remaining users and randomly distribute the relevant components, set as the control group;

[0061] The number of users selected in the experimental group is the same as that in the control group, set as M.

[0062] (3) Conduct data mining and analysis on the click behavior of the target group:

[0063] Calculate the actual number of clicking users: The actual number of clicking users (A) is the sum of the number of clicking users for each component after removing duplicates.

[0064] (4) Quantify the indicators of the control group and the experimental group:

[0065] Accuracy rate (P) = number of clicking users / number of distributed users = A / M;

[0066] Calculate the click-through rate PR of the number of users with label T in different groups, where T is several inferred labels in step (1),

[0067] PR = number of clicking users / number of distributed users = A / M;

[0068] Gain of the number of label users = PR (experimental group) - PR (control group) = A (experimental group) / M - A (control group) / M = (A (experimental group) - A (control group)) / M

[0069] Output the importance of the label, more precisely:

[0070] Importance of the label = gain of the label / accuracy rate = (A (experimental group) - A (control group)) / (M * (A / M)) = (A (experimental group) - A (control group)) / A;

[0071] Taking the elderly label as an example, then A (experimental group) is the number of users who click on the retiree component, which is the same as the actual number of users A who click on the retiree component.

[0072] Therefore, the above formula = 1 - A (control group) / A

[0073] If the number of users who click on the "retired group" component in the experimental group is larger, then the ratio of A (control group) / A is higher, which indicates that this label is less important, that is, even without this label, users will still click; if the ratio of A (control group) / A is lower, it shows that this label is very important.

[0074] (5) Conduct portrait label evaluation:

[0075] Based on the importance and accuracy of the label, infer whether this label is important for this business, that is, whether it is necessary to continuously maintain this label, thereby reducing the maintenance workload and using more energy to maintain labels with stronger importance. It is of great significance for the subsequent continuous update of labels.

[0076] The above-described embodiments only represent the specific implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. A user portrait label evaluation method, characterized in that Including the following steps: Step 1: Construct intelligent creative components according to the tags to be evaluated; Step 2: Use the automated bottoming engine to distribute the intelligent creative components, complete the distribution of the portrait posterior components of the target group, and obtain the experimental group and the control group; Step 3: Conduct data mining on the click behaviors of the target group and calculate the actual number of clicking users A; Step 4: Quantify the indicators of the experimental group and the control group to obtain the accuracy and importance of the tags; In Step 4, the quantification of the indicators of the experimental group and the control group specifically includes: Calculate the click accuracy rate P: P = actual number of clicking users / (actual number of clicking users + number of misinferred clicking users) = actual number of clicking users / number of distributed users = A / M; Calculate the click-through rate PR of the inferred relevant tags among different groups of users: PR = actual number of clicking users / number of distributed users = A / M; Calculate the tag user number gain: tag user number gain = PR(experimental group) - PR(control group) = A(experimental group) / M - A(control group) / M = (A(experimental group) - A(control group)) / M; Calculate the tag importance S: S = tag user number gain / accuracy rate = (A(experimental group) - A(control group)) / (M * (A / M)) = (A(experimental group) - A(control group)) / A; Since A(experimental group) is the number of users who actually click on the inferred tag and is consistent with the actual number of clicking users A, the tag importance S can be simplified as: S = 1 - A(control group) / A. If A(control group) / A is higher, it means that this tag is less important, and even if this tag is not inferred, users will still click; Step 5: Conduct portrait tag evaluation based on the accuracy and importance of the tags, and infer whether this tag is important for the corresponding business, that is, whether it is necessary to continuously maintain this tag.

2. The user portrait label evaluation method according to claim 1, characterized in that Step 1 specifically includes: Infer the relevant tags of the tag according to the user behavior data and requirements; Develop relevant intelligent creative components according to the relevant tags to attract users.

3. The user portrait label evaluation method according to claim 2, wherein Step 2 specifically includes: Distribute the intelligent creative components to each intelligent terminal according to the requirements for users to click, and collect relevant click data; For the inferred relevant tags, conduct personalized user screening respectively, place the inferred relevant tags at the top, and randomly distribute the other tags in order, which is set as the experimental group; Select a part of users from the remaining users and randomly distribute the relevant components, which is set as the control group.

4. The user portrait label evaluation method according to claim 3, wherein The number of users selected for the experimental group and the control group is the same, which is set as M.

5. The user portrait label evaluation method according to claim 4, characterized in that In Step 3, the actual number of clicking users A is the deduplicated sum of the number of clicking users for each component.

Images