An individualized recommendation method based on automobile field tags

Abstract

The application discloses a personalized recommendation method based on tags in the automobile field, which establishes the relationship between the user and the knowledge through the tags, that is, the user behavior information is acquired from a database, the user behavior information in a period of time is selected according to time filtering, and the user behavior information is subjected to data processing to remove abnormal behavior information; in combination with time attenuation, the user's browsing, downloading and collecting behaviors on the knowledge are counted to obtain the statistical score of the user on the knowledge; subsequently, the user is added with the behavior tags according to the statistical score of the user's behaviors on the knowledge; in combination with the self-defined tags selected by the user and the behavior tags obtained according to the user's behaviors, the user is recommended with the related knowledge under the tags, and the knowledge is sorted according to time and the statistical score to select the first several data for recommendation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of personalized recommendation, in particular to a personalized recommendation method based on automobile field tags. BACKGROUND

[0002] With the progress of science and technology, the related requirements in the field of recommendation system are increasingly complex, the tasks to be completed are more difficult, and the effects to be met are increasingly improved. Among them, the recommendation method based on automobile field tags is a representative. The automobile field tags are abstract representations of professional terms in dimensions such as processing technology, part model, vehicle model name, task type and department type in automobile design and production manufacturing. It contains a set of special names in the automobile field such as automobile design and production manufacturing and test processing. The traditional recommendation algorithm based on tags only judges whether the user has selected the item, and does not pay attention to the research on user behavior information, so that the tag information cannot be effectively and reasonably applied.

[0003] At present, the personalized recommendation method based on automobile field tags is less studied, and in actual production practice, due to the complexity of professional knowledge, automobile field personnel cannot obtain the required professional knowledge in time, therefore, the application proposes a method for personalized recommendation by applying automobile field tags. SUMMARY

[0004] To solve the above technical problems, the application provides a personalized recommendation method based on automobile field tags, which is based on five-dimensional tags of knowledge classification, task, department, part and vehicle model in the automobile field, and combines hot recommendation and user-based collaborative filtering algorithm to recommend the interested knowledge for automobile field professionals, which has reference significance for automobile field personalized recommendation.

[0005] The purpose of the application is achieved by the following technical solutions:

[0006] A personalized recommendation method based on automobile field tags, comprising:

[0007] A. Formulating the dimension information of the automobile field as the automobile field tags according to the business;

[0008] B. Counting the tags and selecting part of the tags as the behavior tags of the user;

[0009] C. According to the behavior record of the user and in combination with the time weight decay function, counting the click, download and collection amount of the user on the knowledge, accumulating and summing the statistical amount according to the related weight to obtain the behavior statistical score of the user on the knowledge;

[0010] D, judge whether the user exists a label, yes, find the knowledge under the label, otherwise, solve the user cold start problem, sum the user knowledge behavior result to obtain the score of the knowledge;

[0011] E, sort the statistical weighted knowledge according to time and weight result in descending order, and make difference set with the user behavior record in the recent period of time;

[0012] F, select the user behavior record in the recent period of time, and perform statistical calculation according to step C to obtain the statistical score of the knowledge, and sort the top several knowledge in descending order according to the score;

[0013] G, according to the behavior score of the user and the knowledge, and the correlation between the users, the top several knowledge are recommended for the user by using the user-based collaborative filtering algorithm;

[0014] H, add the recommended several results to the recommendation result set, take the intersection of the label-based recommendation result and the collaborative filtering recommendation result, then sort the intersection result in descending order according to time, and add the top several data from the remaining label recommendation result and collaborative filtering recommendation result to the recommendation result set.

[0015] Compared with the prior art, one or more embodiments of the present application can have the following advantages:

[0016] The present application establishes a label system in the automobile field according to the knowledge classification, department, parts, task and vehicle type and other business dimensions in the automobile field, and combines the automobile field label, hot data and user-based collaborative filtering algorithm to perform personalized recommendation, solves the user cold start problem, and better discovers the content interested by the user. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a personalized recommendation method flowchart based on the automobile field label;

[0018] Figure 2 is a time decay function diagram. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with embodiments and drawings.

[0020] As shown in the figure, it is a personalized recommendation method flowchart based on the automobile field label, which includes the following steps: Figure 1

[0021] Step 10 establishes a set of automobile field labels according to the five dimensions of knowledge classification, department, task, parts and molding in the automobile field according to the business.​

[0022] Step 20, the user can customize the selected label as the user's custom label; meanwhile, the user's behavior information in a period of time (according to the user's behavior in the current system) can be obtained from the database, and different weights are given to each user behavior record according to the user's click, download, collection and other behaviors on knowledge. Then, according to the time of the user behavior record, time decay is applied to the weight of each user record, so that the behavior closer to the current has a higher weight.

[0023] For each behavior record, all knowledge labels in each dimension have the same weight, so the weight sum of each label in each dimension is obtained by respectively summing all labels in each dimension in all behaviors. Finally, the labels in each dimension are sorted in descending order, and the top several labels in each dimension are selected.

[0024] Step 30, according to the user's behavior record, the user's click, download and collection of knowledge can be counted, and with the passage of time, the weight of the click, download and collection of knowledge in the behavior record will gradually decrease, that is, the longer the time, the smaller the weight of the behavior, and the weight score in each behavior record is respectively accumulated and summed by combining time decay, to obtain the behavior statistics of knowledge.

[0025] Step 40, if the user has a label, find the knowledge under the label, and calculate the score of the knowledge according to the click, download and collection of knowledge. If the user has no label, use the solution method of the user cold start problem.

[0026] Step 50, the weighted knowledge is sorted in descending order according to the time and weight results, and the difference set is made with the user's behavior record in a period of time (such as a week, according to the specific situation), the purpose is to remove the knowledge with behavior in the user's behavior record in a period of time, and select the top several results.

[0027] Step 60, select the user's behavior record in the recent period of time (such as one day), and calculate the statistical score of the knowledge according to the user's behavior statistics of the knowledge in step 30, and sort the results in descending order according to the score, and select the top several knowledge.

[0028] Step 70, according to the user's behavior score of the knowledge obtained in step 30, and because the behavior between users in the department has strong correlation, the user-based collaborative filtering algorithm is used to recommend the user, and the top several knowledge is selected.

[0029] Step 80 adds several hot recommended results to the recommended result set; takes the intersection of the tag-based recommended results and the collaborative filtering recommended results to ensure that the knowledge conforms to the user's interests, then sorts the intersection results in reverse order according to time to ensure the novelty and diversity of the recommended results; subsequently, selects the first several data from the remaining tag recommended results and collaborative filtering recommended results and appends them to the recommended result set.

[0030] In step 20, the time decay function is shown in formula (1):

[0031] N(t) = N0e -αt (1)

[0032] Wherein, a > 0 is an exponential decay constant.

[0033] Wherein, N(t) is the value of N at time t, N0 = N(0) is the initial value of N at time 0, which is also the maximum value when t ≥ 0, and t represents the number of days.

[0034] Suppose the initial hotness of a piece of knowledge is N0, then its hotness decreases in an exponential form with the passage of time, and the calculation formula is formula (1).

[0035] Suppose N_0 = 1, then the hotness of the knowledge changes with time as shown in the following formula (2): Figure 2

[0036] In step 40, the user cold start problem refers to the case where the current user is a new user and cannot be recommended according to the user tags. The solution is to find all the knowledge in the user behavior in the department and perform statistical weighting; if the knowledge amount is still small, other knowledge in the department can be obtained to fill in.

[0037] In step 70, the user correlation recommendation is performed using the user-based collaborative filtering algorithm, and the principle is:

[0038] 1. Find a set of users with similar interests

[0039] To find a group of users with similar interests, we first need to define an index to measure the similarity between users. The commonly used ones are the Jaccard formula and the cosine similarity, which calculate the similarity between users.

[0040] Suppose N(a) and N(b) are the sets of items that users like.

[0041] Then according to the Jaccard formula, the similarity between users is calculated as shown in formula (2):

[0042]

[0043] ​Another cosine similarity calculation method (such as formula (3)) is:

[0044]

[0045] Example: Assuming A-E represent 5 users, a-f represent 6 sample items, and the details of the items that users like are as shown in Table 1:

[0046] Table 1

[0047] A a, b, d B c, d C a, b, e D a, c, f E b, e, f

[0048] 1.1 Build an item-user information table

[0049] The above data needs to be converted first, converted into item-user details, as shown in Table 2:

[0050] Table 2

[0051] a A, C, D b A, C c B, D d A, B e C, E f D, E

[0052] 1.2 Build a user similarity matrix

[0053] Then, for the same item that the user likes, mark 1, for example, users C, E like the same item, then mark the corresponding position in the matrix as 1. As shown in Table 3:

[0054] Table 3

[0055]

[0056]

[0057] 1.3 Calculate user similarity

[0058] According to the cosine similarity, the complete similarity is obtained as shown in Table 4:

[0059] Table 4

[0060] A B C D E A 0 0.25 0.5 0.2 0 B 0.25 0 0 0.25 0 C 0.5 0 0 0.2 0.2 D 0.2 0.25 0.2 0 0.2 E 0 0 0.2 0.2 0

[0061] 2. For the target user, make item recommendations

[0062] According to the user-item similarity matrix, make recommendations, assuming that user s has k similar users, marked as T(s, k), and extract all items in T, and remove the items that user s has liked. Then, for each candidate item i, define the interest of user s in candidate item i, as shown in formula (4):

[0063] p(s,i)=∑ v∈T(s,k)∩N(i) w sv ×r vi (4)

[0064] wherein r vi represents the like degree of the user v in the user group T to the item i, and defaults to 1. If the user gives specific score data, the like degree of the user to the item can be more accurately described.

[0065] Although the embodiments of the present application are disclosed as above, the content described is only for the purpose of facilitating understanding of the present application, and is not intended to limit the present application. Any person skilled in the art of the present application can make any modification and change in the implementation form and details without departing from the spirit and scope of the present application, but the patent protection scope of the present application shall be subject to the scope defined by the appended claims.

Claims

1. A method of personalized recommendation based on tags in the automotive field, characterized in that, The method comprises the following steps: Step 10: form the dimension information in the automobile field as the automobile field label according to the business; Step 20: count the label, and select part of the label as the behavior label of the user; Step 30: according to the behavior record of the user, and in combination with the time weight decay function, count the click, download and collection amount of the user to the knowledge, add the statistical amount according to the related weight to obtain the behavior statistical score of the user to the knowledge; Step 40: judge whether the user exists the label, if yes, find the knowledge under the label, otherwise, solve the user cold start problem, add the knowledge behavior result to obtain the score of the knowledge; Step 50: sort the statistical weighted knowledge in the descending order according to the time and weight result, and make a difference set with the behavior record of the user in the recent period of time, remove the knowledge with the behavior of the user in the recent period of time, and select the first several pieces of knowledge; Step 60: select the behavior record of the user in the recent period of time, and obtain the statistical score of the knowledge according to the statistical calculation in step 30, sort the first several pieces of knowledge in the descending order according to the score; Step 70: according to the behavior of the user and the behavior score of the knowledge, and the correlation between the users, use the user-based collaborative filtering algorithm to recommend the first several pieces of knowledge for the user; Step 80: add the recommended several pieces of results to the recommendation result set, take the intersection of the label-based recommendation result and the collaborative filtering recommendation result, then sort the intersection result in the descending order according to the time, and select the first several pieces of data from the remaining label recommendation result and the collaborative filtering recommendation result and add to the recommendation result set. 2.The car field tag-based personalized recommendation method according to claim 1, wherein, The dimension information in step 10 comprises the knowledge classification, department, spare part, task and vehicle type according to the business. 3.The car field tag-based personalized recommendation method according to claim 1, wherein, The behavior label in step 20 comprises that the user selects the label as the user-defined label, and the user obtains the behavior information of the user in the recent period of time from the database, and according to the click, download and collection behavior of the user to the knowledge, gives different weights to each user behavior record, and then according to the time of the user behavior record, applies time decay to the weight of each user record, so that the behavior closer to the current has a higher weight. 4.The car field tag-based personalized recommendation method according to claim 3, characterized in that, The time decay function is shown in formula (1): N(t) = N0e -αt (1) Wherein, alpha>0 is called exponential decay constant; Wherein, N(t) is the value of N at t time, N0=N(0) is the initial value of N at 0 time, which is also the maximum value when t>0, and t represents the number of days. 5.The car field tag-based personalized recommendation method according to claim 1, wherein, The user cold start problem refers to the case that the current user is a new user and cannot be recommended according to the user label; the solution to this case is to find the knowledge in the behavior of all users in the department and statistically weight. 6.The car field tag-based personalized recommendation method according to claim 1, wherein, In step 70: the user-based collaborative filtering algorithm is used to recommend the correlation of the user, which comprises: 1) find the user set with similar interests; establish the item-user inverted list; establish the user similarity matrix; calculate the user similarity; 2) recommend the item for the target user.

Images