CN111695023A - Information recommendation method and device, storage medium and equipment - Google Patents
Information recommendation method and device, storage medium and equipment Download PDFInfo
- Publication number
- CN111695023A CN111695023A CN201910180979.8A CN201910180979A CN111695023A CN 111695023 A CN111695023 A CN 111695023A CN 201910180979 A CN201910180979 A CN 201910180979A CN 111695023 A CN111695023 A CN 111695023A
- Authority
- CN
- China
- Prior art keywords
- user
- preset
- historical behavior
- behavior information
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Abstract
The embodiment of the invention discloses an information recommendation method, an information recommendation device, a storage medium and equipment. The method comprises the following steps: determining target users corresponding to the current users, wherein the target users comprise users with similarity meeting preset requirements with the current users; acquiring first historical behavior information corresponding to a current user and second historical behavior information corresponding to a target user; determining an object which appears in the second historical behavior information but does not appear in the first historical behavior information as an object to be recommended; and recommending the relevant information of the object to be recommended to the current user. By adopting the technical scheme, the embodiment of the invention can achieve the technical effect of improving the diversity and novelty of the recommendation information while giving consideration to the recommendation accuracy.
Description
Technical Field
The embodiment of the invention relates to the technical field of computers, in particular to an information recommendation method, an information recommendation device, a storage medium and information recommendation equipment.
Background
The recommendation algorithm is an algorithm in the computer profession, and what the user may like is presumed through some mathematical algorithms. At present, recommendation algorithms are widely applied to network platforms and the like, and some behaviors of users can be utilized to conjecture things which the users may like through some mathematical algorithms.
In the process of implementing the invention, the inventor finds that at least the following technical problems exist in the prior art: current recommendation algorithms can be broadly classified into the following categories: content-based recommendations, association rule-based recommendations, utility-based recommendations, knowledge-based recommendations, and combinatorial recommendations, among others. The recommendation algorithms mainly research the aspects of recommendation precision, time complexity, space complexity, small samples, generalization capability, cold start and the like. However, while the above characteristics are improved, some disadvantages are also brought, so that when the recommendation algorithm is used for information recommendation, the actual requirements of the user cannot be met, and thus the existing information recommendation scheme is still not perfect and needs to be improved.
Disclosure of Invention
The embodiment of the invention provides an information recommendation method, an information recommendation device, a storage medium and information recommendation equipment, which can optimize the existing information recommendation scheme.
In a first aspect, an embodiment of the present invention provides an information recommendation method, including:
determining target users corresponding to current users, wherein the target users comprise users with similarity to the current users meeting preset requirements;
acquiring first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
determining an object which appears in the second historical behavior information but does not appear in the first historical behavior information as an object to be recommended;
and recommending the relevant information of the object to be recommended to the current user.
In a second aspect, an embodiment of the present invention provides an information recommendation apparatus, including:
the target user determining module is used for determining a target user corresponding to a current user, wherein the target user comprises a user of which the similarity with the current user meets a preset requirement;
the historical behavior information acquisition module is used for acquiring first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
the object to be recommended determining module is used for determining the objects which appear in the second historical behavior information but do not appear in the first historical behavior information as the objects to be recommended;
and the information recommending module is used for recommending the relevant information of the object to be recommended to the current user.
In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to implement the information recommendation method according to the embodiment of the present invention.
In a fourth aspect, an embodiment of the present invention provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the information recommendation method according to an embodiment of the present invention.
According to the information recommendation scheme provided by the embodiment of the invention, the target user corresponding to the current user is determined, the target user comprises a user of which the similarity with the current user meets the preset requirement, first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user are respectively obtained, an object appearing in the second historical behavior information but not appearing in the first historical behavior information is determined as an object to be recommended, and relevant information of the object to be recommended is recommended to the current user. By adopting the technical scheme, the target user is a user having certain commonality with the current user, and the information which is related to the target user and is not related to the current user is recommended to the current user, so that the technical effect of improving the diversity and novelty of the recommended information while considering the recommendation accuracy can be achieved.
Drawings
Fig. 1 is a schematic flowchart of an information recommendation method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of another information recommendation method according to an embodiment of the present invention;
fig. 3 is a schematic flowchart of another information recommendation method according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a feature construction provided by an embodiment of the present invention;
fig. 5 is a schematic diagram of a KNN principle according to an embodiment of the present invention;
FIG. 6 is a diagram illustrating classification near the boundary of an SVM according to an embodiment of the present invention;
fig. 7 is a schematic diagram of a scheme of an information recommendation method according to an embodiment of the present invention;
fig. 8 is a block diagram of an information recommendation apparatus according to an embodiment of the present invention;
fig. 9 is a block diagram of a computer device according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained by the specific implementation mode in combination with the attached drawings. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.
Before discussing exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart may describe the steps as a sequential process, many of the steps can be performed in parallel, concurrently or simultaneously. In addition, the order of the steps may be rearranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figure. The processes may correspond to methods, functions, procedures, subroutines, and the like.
In the related art, the following recommendation schemes are mainly adopted when information recommendation is performed:
(1) content-based recommendations
The theoretical basis of the content-based information recommendation method mainly comes from information retrieval and information filtering, and the so-called content-based recommendation method is to recommend recommended items which are not touched by a user to the user according to the past browsing records of the user.
(2) Association rule based recommendations
The recommendation based on the association rule is based on the association rule, the purchased commodity is used as a rule head, and the rule body is used as a recommendation object. Association rule mining can discover the relevance of different commodities in the sale process, and has been successfully applied in the retail industry. The management rule is that a transaction database counts the proportion of transactions purchasing the commodity set X and purchases the commodity set Y at the same time, and the intuitive meaning is that the user has a tendency to purchase some commodities to purchase other commodities. Such as milk, while many people buy bread at the same time.
(3) Utility-based recommendations
The utility-based recommendation is calculated based on the utility of the user use item, and the core problem is how to create a utility function for each user, so that the user profile model is largely determined by the utility function adopted by the system. The benefit of utility-based recommendations is that it takes into account non-product attributes (e.g., reliability of the provider) and product availability, etc. into the utility calculation.
(4) Knowledge-based recommendations
Knowledge-based recommendations can be viewed to some extent as a reasoning technique that is not based on recommendations based on user needs and preferences. Knowledge-based methods differ significantly in the functional knowledge they use. Utility knowledge is knowledge of how an item satisfies a particular user, and therefore explains the relationship of needs and recommendations, so the user profile can be any inference-capable knowledge structure, either a query that the user has normalized, or a more detailed representation of the user's needs.
(5) Combined recommendations
In practice, combined recommendations are often used because of the advantages and disadvantages of each recommendation method. Most studied and applied are combinations of content recommendations and collaborative filtering recommendations. The simplest approach is to use a content-based approach and a collaborative filtering recommendation approach to generate a recommendation prediction result, and then combine the results in some way. Although there are many methods for recommending combinations in theory, it is not always effective in a specific problem, and one of the most important principles for recommending combinations is to avoid or compensate the weaknesses of the respective recommending technologies by combining.
Based on the above, the main research angles in the related art include recommendation accuracy, time complexity, space complexity, small samples, generalization capability, cold start, and the like. While improving the above characteristics, some disadvantages are brought, and the inventors found that neither diversity nor novelty index is considered in the above recommendations. For example, in an e-commerce website, a user may want to buy a notebook computer recently, so the user browses the computer brands and then the user purchases a notebook computer of a certain brand. The recommendation list of the e-commerce website is a behavior of the user, the user can recommend notebook computers of other brands, and some recommended and better e-commerce websites can recommend related commodities such as a keyboard, a mouse, an earphone and the like. However, for the user, the recommended notebook, keyboard, mouse, etc. are all within the user's expectation, but may not be purchased for some reason, such as funds or waiting for new money. Such a recommendation function may not be very useful for some users, and is not intended for a recommendation algorithm, and can only be used as a marketing means, such as a recommendation reminder for a certain brand of activity. For the recommendation information, a user may prefer to buy a notebook computer and recommend a watch to the notebook computer, and the two are not strongly related, but have diversity and novelty. In the embodiment of the invention, the inventor obtains a corresponding technical scheme from the viewpoints of diversity and novelty.
Fig. 1 is a flowchart illustrating an information recommendation method according to an embodiment of the present invention, where the method may be executed by an information recommendation apparatus, where the apparatus may be implemented by software and/or hardware, and may be generally integrated in a computer device such as a server. As shown in fig. 1, the method includes:
For example, the technical solution of the embodiment of the present invention may be applied to information recommendation in various fields, and is not particularly limited. For example, the method can be applied to commodity information recommendation in an e-commerce platform, recommendation of multimedia contents such as songs, movies and pictures in a multimedia platform, course information recommendation or topic recommendation in an online education platform, and the like. Accordingly, the computer device in the embodiment of the present invention may be a server corresponding to the network platform. The users may be people facing the network platform, such as consumers of the e-commerce platform, listeners or viewers of the multimedia platform, students of the online education platform, and so on.
Exemplarily, the similarity between the current user and other users can be calculated in advance, corresponding target users are screened out for recording, and when the step is executed, the target users are determined in a mode of directly reading the records, so that the advantages of improving the efficiency of determining the target users and reducing the operation burden of the server are achieved; in addition, when the step is executed, the similarity between the current user and other users is calculated, and the target user is determined according to the calculation result, so that the advantage that the related information for calculating the similarity is up-to-date is ensured, and the target user is determined more accurately. Specifically, the selection can be performed according to factors such as configuration of the server, computing capacity, requirements of application scenarios, and the like. The other users may be all users except the current user in the platform, or may be a set of users screened according to a certain rule.
For example, when the similarity is calculated, the indexes or characteristics may be set according to actual situations, such as user attributes (including age, gender, area, occupation, and the like), historical behavior characteristics (such as categories of goods purchased once, types of songs listened to once, or courses attended once, and the like), and other information, and the embodiment of the present invention is not limited in particular.
And 102, acquiring first historical behavior information corresponding to a current user and second historical behavior information corresponding to a target user.
For example, the historical behavior information may include some behavior records generated by the user during the use process, and the specific content is related to the applied network platform. Taking the e-commerce platform as an example, the record of the user purchasing the goods, the evaluation information of the user on the purchased goods, the information of the goods which the user has paid attention to or collected, the information of the goods which the user joins in a shopping cart, and the like can be included. Taking the multimedia platform as an example, the multimedia content records watched or listened to by the user, the evaluation information of the multimedia content by the user, the multimedia content information once watched or collected by the user, the multimedia content information once purchased, and the like can be included.
And step 103, determining the object which appears in the second historical behavior information but does not appear in the first historical behavior information as the object to be recommended.
Illustratively, the object to be recommended is related to the applied network platform. Taking an e-commerce platform as an example, the object to be recommended can be a commodity; taking a multimedia platform as an example, the object to be recommended can be multimedia contents such as songs, movies, pictures and the like; taking the online education platform as an example, the object to be recommended may be a course or a topic, etc.
For example, a first object set related to the first historical behavior information may be extracted from the first historical behavior information, a second object set related to the second historical behavior information may be extracted from the second historical behavior information, an intersection of the first object set and the second object set is obtained, and the obtained intersection is subtracted from the second object set, so as to obtain the object to be recommended. Because the target user has certain similarity with the current user and the historical behavior of the target user has certain randomness, the target user has diversity and novelty on the basis of ensuring that the object to be recommended meets the requirements of the current user to a certain extent.
And 104, recommending the relevant information of the object to be recommended to the current user.
For example, the related information of the object to be recommended may include a name, a profile, a representative image, a price, a purchase link, and the like of the object to be recommended, and may be determined according to attributes of the object to be recommended, a policy of a network platform, and the like. Generally, the server can push the relevant information of the object to be recommended to the corresponding network platform client, so as to instruct the client to provide the information to the current user through a screen display or the like.
The information recommendation method provided by the embodiment of the invention determines a target user corresponding to a current user, wherein the target user comprises a user the similarity of which to the current user meets a preset requirement, respectively obtains first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user, determines an object appearing in the second historical behavior information but not appearing in the first historical behavior information as an object to be recommended, and recommends related information of the object to be recommended to the current user. By adopting the technical scheme, the target user is a user having certain commonality with the current user, and the information which is related to the target user and is not related to the current user is recommended to the current user, so that the technical effect of improving the diversity and novelty of the recommended information while considering the recommendation accuracy can be achieved.
In some embodiments, the determining the target user corresponding to the current user includes: respectively calculating the similarity between the current user and each user in a preset user set; and determining the users with the similarity meeting the preset requirement as target users. The method has the advantages that the preset user set is set in advance, the calculation amount in similarity calculation is reduced, and meanwhile the target user can be determined more accurately. The preset user set may be filtered according to user attributes or user behaviors, and the specific filtering manner is not limited in the embodiments of the present invention. The preset requirement may be, for example, that the similarity value reaches a similarity threshold.
In some embodiments, before separately calculating the similarity between the current user and each user in the preset user set, the method may further include: measuring each user in the designated user group according to a preset measurement index; and generating a preset user set according to the users with the weighing results meeting the preset conditions. The advantage of this arrangement is that the preset set of users can be more reasonably determined. Generally, users are more reliant on the objects they recommend by influential people, and are more willing to try the objects they recommend because they are trusted. However, influential people generally have certain representativeness and are more active on a network platform, so that some indexes related to the activity degree can be referred to determine preset measurement indexes for measuring each user in a specified user group, and certainly, other factors can be referred to select the preset measurement indexes, and the embodiment of the invention is not limited. The designated user group may be all users in the network platform, or may also be a user group that is simply screened, such as a user who is screened according to an age group, or a user who has a friend relationship with the current user on the network platform, and the like. The preset condition may be, for example, that the preset metric reaches a corresponding threshold.
Fig. 2 is a schematic flow chart of another information recommendation method provided in an embodiment of the present invention, which is described by taking an application scenario of an e-commerce platform as an example, and specifically, the method includes the following steps:
For example, the designated user group may be all users in the e-commerce platform, or may be users having a friend relationship with the current user.
Illustratively, when the screened users have shopping diversity, more diversified and novel commodities can be recommended to other users, so that the shopping behavior diversity can be used as a preset measuring index. The diversity of shopping behavior can be measured by calculating how many categories of goods are purchased in a period through big data. The commodity category can be determined according to a classification mode provided in an e-commerce platform, and can be classified into clothes, personal care cleaning, clocks, jewelry, mobile phone numbers, computer office, household appliances, fresh food, furniture home decoration, toy musical instruments and the like. Each major category can be divided more carefully, for example, the fresh food can comprise fruits, meat, seafood, aquatic products, snack foods, vegetable and egg products, grain and oil seasoning, and the like.
Alternatively, the diversity of shopping behavior may be determined according to the following formula:
wherein C is the number of all categories, Categoryu,tNumber of categories of goods purchased by user u in period t, Diversityu,tI.e. the diversity of the user u within the period t. The data can be obtained by writing a Structured Query Language (SQL) for querying through a big data platform.
Illustratively, when the screened users have higher shopping frequency, the purpose of updating the objects to be recommended in time can be achieved, so the frequency of shopping behaviors can be used as a preset measurement index. The frequency of shopping activity can be measured by how many times a good is purchased in a big data computing cycle. The query can be obtained by writing SQL by a big data platform. Alternatively, each order placed may be considered a purchase, and each purchase of an item may also be considered a purchase.
Illustratively, when the evaluation of the screened user on the purchased commodity is effective evaluation, the real feeling of the user after experiencing the commodity can be comprehensively reflected, so that the effective shopping evaluation rate can be used as a preset measurement index. The effective shopping evaluation rate can be obtained by classifying and distinguishing whether an evaluation is an effective evaluation through texts, counting the number of effective evaluations in a period, and dividing the number by the number of commodities purchased in the period. The method can perform semantic recognition and other processing on the content evaluated by the user, and further recognize whether the evaluation of the user is real or not, so that text classification is realized, and the specific classification mode is not limited.
Alternatively, the effective shopping rate may be determined according to the following formula:
wherein, Pu,tFor all the goods purchased by the user during the u period t, Eu,tThe number of effective evaluations, EE, in the user u period tu,tNamely the effective evaluation rate in the user u period t. Pu,tCan be obtained by compiling SQL by a big data platform for query, Eu,tWhether the evaluation is effective evaluation can be distinguished through a text classification technology, and then the number of the effective evaluation is counted.
For example, some users may use the screening vulnerability to become one member of a preset user set in order to satisfy personal interests, and further affect other normal users to achieve personal purposes, and these users are called attack users. The attack user will interfere with the ordinary user, so whether the attack detection is passed or not can be used as a preset measuring index. Whether the preset user set is an attack user or not can be judged when the preset user set is screened, and judgment of detecting the attack user can be filtered through a classifier, and the specific implementation mode is not limited in the embodiment of the invention.
For example, a corresponding metric may be specified for each preset metric, and when the corresponding metric is satisfied, it is determined that the preset condition is satisfied. For example, corresponding thresholds, such as a diversity threshold, a frequency threshold, and an evaluation rate threshold, may be set for the diversity of shopping behaviors, the frequency of shopping behaviors, and the effective shopping evaluation rate, respectively. For attack detection, the pass attack detection is set as the corresponding metric.
Illustratively, the similarity between the current user and each user in the preset user set based on the user attribute and/or the shopping behavior is calculated respectively. The user attributes may include, for example, age, gender, location, occupation, etc., and the shopping behavior may include, for example, categories of commodities purchased once, brands of commodities purchased once, specific commodities purchased, number of commodities purchased once, and whether to perform evaluation after purchasing a commodity. When the similarity calculation is performed, a corresponding similarity calculation mode may be selected according to the referred index, and the embodiment of the present invention is not particularly limited. For example, it may be to calculate a euclidean distance, a manhattan distance, a cosine similarity or a pearson correlation coefficient, or the like.
And step 204, determining the user with the similarity meeting the preset requirement as a target user.
For example, a similarity threshold may be set, and the user who reaches the threshold may be determined as the target user. The number of target users is not limited in the embodiment of the present invention, and may be 1 or more. Optionally, the users may be ranked according to the similarity, and a preset number of users with higher similarity are determined as the target users. The preset number can be set according to actual conditions, such as 3.
And step 206, determining the commodities which appear in the second historical shopping behavior information but do not appear in the first historical shopping behavior information as objects to be recommended.
The number of the objects to be recommended is not limited, and the number can be set according to the page layout of the e-commerce platform and the like. Alternatively, the goods that are purchased by the target user but not purchased by the current user may be determined as the objects to be recommended.
For example, since the difference between the commodities in the specific shopping behaviors of the current user and the target user may be large, the objects to be recommended are more, and therefore, the commodities with higher sales volume or higher rating rate can be screened as the objects to be recommended according to a certain rule, such as sorting according to sales volume or rating rate.
And step 207, recommending the relevant information of the object to be recommended to the current user.
For example, the object to be recommended, that is, the related information of the item to be recommended may be determined according to the page layout of the e-commerce platform, and the like. For example, the name of the commodity, the schematic diagram of the commodity, the price and the like can be included.
The information recommendation method provided by the embodiment of the invention takes an application scene of an e-commerce platform as an example, firstly determines a preset user set with influence according to the diversity of shopping behaviors, the frequency of the shopping behaviors, the effective shopping evaluation rate and whether attack detection passes through preset measurement indexes and the like, then screens out a target user from the preset user set according to the similarity with the current user, determines commodities purchased by the target user and not purchased by the current user as objects to be recommended, and recommends related information to the current user.
Fig. 3 is a schematic flow chart of another information recommendation method according to an embodiment of the present invention, which is described by taking an application scenario of an e-commerce platform as an example, and specifically, the method includes the following steps:
For example, the model calculation period may be set according to the calculation capability of the server and the application scenario, for example, the model calculation period is 1 week, and when the current time reaches the time corresponding to the model calculation period, the model update event is triggered.
The preset recommendation model can mainly complete the following work: generating a preset user set; when receiving input relevant information of a current user, calculating the similarity between the current user and each user in a preset user set, and further screening out a target user; and obtaining a list of objects to be recommended according to the shopping history conditions of the current user and the target user.
When the model is updated, the preset user set may be mainly regenerated, and since the shopping behavior of the user is updated in real time, the preset user set may change in a new period and needs to be regenerated. Illustratively, each user in a designated user group is measured according to preset measurement indexes, wherein the preset measurement indexes comprise diversity of shopping behaviors, frequency of the shopping behaviors, effective shopping evaluation rate and whether attack detection passes or not. Wherein the designated user group may be all users in the e-commerce platform.
The calculation of the diversity of shopping behavior, frequency of shopping behavior, and effective shopping rate may refer to the related contents provided above. An alternative attack detection approach is provided below as an illustrative illustration.
Constructing a feature vector of a first user based on shopping evaluation information of the first user; inputting the feature Vector into a preset classifier, and determining whether the first user passes attack detection or not according to an output result of the preset classifier, wherein the preset classifier is generated based on a Support Vector Machine (SVM) and a K-nearest neighbor (KNN) algorithm. This has the advantage that it can be detected accurately whether a user is an attacking user.
The attacking user will typically develop the strategy used by the attack profile based on knowledge about the recommendation system, scoring database, project, and user. The attack models mainly adopted at present include average attack, random attack, popular attack, segmented attack and the like, and the attack of the user profile can be divided into types of push attack, nuclear attack and the like according to different attack purposes. Specifically, in the embodiment of the present invention, the score data of each user may be analyzed, and some evaluation indexes are set by using statistics to distinguish an attack user from a normal user, and whether the evaluation indexes are valid or not may be verified by using real data. And then, converting the user evaluation matrix into an evaluation index matrix, and constructing a feature vector of each user. Fig. 4 is a schematic diagram of feature construction provided by an embodiment of the present invention, as shown in fig. 4, a User evaluation matrix is on the left side, and an evaluation index matrix is on the right side, where User represents a User, Item represents an Item, r represents an evaluation value, Eva represents an evaluation index, and v represents a value calculated by the evaluation index. After the matrix conversion is completed, the feature vector corresponding to each user can be constructed according to v in the evaluation index matrix. When the evaluation index includes a plurality of values, v includes a plurality of values, respectively.
Illustratively, in the embodiment of the present invention, the preset classifier is obtained by combining the SVM and the KNN. The SVM can be used as a basic classifier, and the KNN can be used as an auxiliary classifier for improving the classification accuracy of points near the SVM interface. The SVM controls the hyperplane interval to solve the over-fitting problem of the function, the dimension problem is solved through the kernel function, the Structure Risk Minimization (SRM) principle is better realized, and the method has good popularization capability. When the sample point is near the interface, the SVM classification effect is not ideal enough, and KNN is adopted for assistance in the embodiment of the invention. The KNN core idea is to find K nearest neighbors and judge the category of the sample to be classified according to the neighbors. Fig. 5 is a schematic diagram of KNN principle provided by an embodiment of the present invention, as shown in fig. 5, a triangle represents an unknown shape, and when K is 4, 4 nearest neighbors are selected, wherein there are 3 circles and 1 square, that is, the unknown shape should belong to a circle. When K is 7, 7 nearest neighbors are chosen, of which 3 circles and 4 squares, i.e. the unknown shape should belong to a square. As can be seen from the graph analysis, classification of KNN is only related to the neighbors, the number of the neighbors is limited, and therefore, the selection of K and whether the data set is balanced have a great influence on the final classification effect of the KNN.
In the embodiment of the invention, the SVM-KNN based attack detection method takes an SVM as a main part and KNN as an auxiliary part. The SVM is used as an initial classifier, experimental analysis shows that the sample points which are wrongly classified by the SVM are basically near the interface, the classification accuracy of the SVM can be improved from the aspect of the sample points which are wrongly classified, and the wrongly classified condition is improved aiming at the sample points which are wrongly classified. The SVM is a typical binary classification problem, namely a non-negative type, namely a positive type, two types are respectively represented on two sides of an SVM interface, and the two types can be respectively replaced by a representative point, so that the SVM can be regarded as a 1NN classifier with one representative point in each type, and when a sample point is near the interface, the classification effect is not good. KNN is based on examples, and auxiliary classification is carried out on sample points which are easy to be wrongly classified near the SVM classification interface by using KNN without training in advance.
FIG. 6 is a schematic diagram of classification near the boundary of an SVM provided by an embodiment of the present invention, as shown in FIG. 6, for a sample point x to be classified, first, the distances between x and x-of two boundaries of the SVM boundary are calculated, and a difference operation is performed on the two calculated distances, if the absolute value of the difference is equal to a given threshold, that is, x is farther from the boundary, as shown by x1 and x2 in regions I and II, the classification can be accurately classified, wherein the threshold is the width of the SVM boundary, that is, the distance between x + and x-. If the absolute value of the difference is smaller than the threshold value, i.e. x is closer to the separation interface, such as x3 in region iii in the figure, it is easily misclassified. The absolute value of the difference is not greater than the threshold, and if the distance of a specific separation interface is required to be calculated, the absolute value can be calculated independently. For points in a region III in the graph, the SVM is easy to carry out wrong classification, KNN auxiliary classification is needed, when a sample point is calculated to be in the region, the distance between the sample point to be classified and all other points is calculated through a KNN method, and specific classification is determined through the number of neighbors.
For example, when a user accesses the e-commerce platform by using a client, the client may send a commodity recommendation request to the server, and at this time, a commodity recommendation event in the server is triggered to obtain a latest updated preset recommendation model.
And step 303, acquiring a historical shopping list corresponding to the current user.
For example, the historical shopping list corresponding to the current user may include a list of items purchased by the user in the near future (e.g., within a year, a month, etc.).
For example, the preset recommendation model may obtain the historical shopping lists of the users in the preset user set, and then perform similarity calculation based on the historical shopping lists of the current users and the historical shopping lists of the users in the preset user set. In the embodiment of the invention, novelty and diversity are mainly considered, the user does not pay attention to the grade of a commodity, and the user pays attention to whether the commodity is purchased, so the cosine similarity can be adopted. The formula of cosine calculated distance is a blue distance, and is generally calculated by using the inner product of the feature vectors or the cosine of an included angle theta, and the smaller the included angle of the vectors is, namely, the larger the cosine value is, the higher the similarity is. The formula for calculating the similarity of two eigenvectors using the cosine distance is as follows:
wherein, wikAnd wjk(1. ltoreq. k. ltoreq.p) are respectively samples diAnd djThe weight of the kth characteristic item can be continuously debugged and calculated to obtain the optimal weight; p is the number of feature terms, i.e., the feature vector spatial dimension. The above formula firstly normalizes the length of the feature vector and then calculates the inner product, and the purpose of normalization is to make samples with inconsistent lengths comparable. For example, the characteristic item may be a certain commodity category or a certain specific commodity, and the value of the characteristic item may be whether the user purchased the commodity or not.
After the target user is determined, comparing the historical shopping list of the current user with the historical shopping list of the target user, determining an object to be recommended, forming a list of the object to be recommended and outputting the list.
And 305, acquiring relevant information of the object to be recommended according to the list of the object to be recommended, and recommending the information to the current user.
Fig. 7 is a schematic view of a scheme of an information recommendation method according to an embodiment of the present invention, and the embodiment of the present invention may be further understood with reference to the schematic view. As shown in fig. 7, a normal user is input into the recommendation model to obtain a recommendation list; the recommendation model screens the influence users by formulating rules (such as algorithms and logics) to obtain an influence user group, calculates the similarity between the normal users and the influence users for the normal users, and associates the normal users and the influence users according to the similarity.
The information recommendation method provided by the embodiment of the invention takes an application scene of an e-commerce platform as an example, regularly updates the preset recommendation model, and determines the preset user set with influence according to preset measurement indexes such as the diversity of shopping behaviors, the frequency of the shopping behaviors, the effective shopping evaluation rate and whether attack detection passes or not during model updating. When commodity recommendation is needed for a current user, a latest preset recommendation model is obtained, the similarity between the current user and an influence user is rapidly and accurately calculated by the model in a cosine similarity calculation mode, a target user is determined, commodities which are purchased by the target user but not purchased by the current user are determined as objects to be recommended, and relevant information is recommended to the current user.
In some embodiments, when the model is updated, the preset user set is regenerated, and the target user can be updated for the current user, that is, the process of determining the target user is advanced to the time before the commodity recommendation event is detected to be triggered, so that for one model calculation period, the target user corresponding to the current user can be determined, the calculation process of determining the target user can be saved, and the recommendation efficiency is further improved. Accordingly, step 304 may be replaced with: and inputting the historical shopping list into a preset recommendation model, so that the preset recommendation model determines that the target user who is determined when the model is updated purchases the commodities but the commodities which are not purchased by the current user are determined as objects to be recommended, and outputting a list of the objects to be recommended. The historical shopping list of the current user is acquired after the commodity recommending event is triggered, so that the current user is latest and the condition that the commodities purchased by the current user are recommended does not occur.
In some embodiments, because the embodiments of the present invention mainly focus on improvement of diversity and novelty, and other recommendation schemes also have respective emphasis points, in actual application, a plurality of recommendation blocks may be set in a page of a client, each block corresponding to a different recommendation scheme, where one scheme is the recommendation scheme provided by the embodiments of the present invention.
Fig. 8 is a block diagram of an information recommendation apparatus according to an embodiment of the present invention, where the apparatus may be implemented by software and/or hardware, and may be generally integrated in a computer device such as a server, and may perform information recommendation by executing an information recommendation method. As shown in fig. 8, the apparatus includes:
a target user determining module 801, configured to determine a target user corresponding to a current user, where the target user includes a user whose similarity to the current user meets a preset requirement;
a historical behavior information obtaining module 802, configured to obtain first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
a to-be-recommended object determining module 803, configured to determine an object that appears in the second historical behavior information but does not appear in the first historical behavior information as an object to be recommended;
and the information recommending module 804 is configured to recommend the relevant information of the object to be recommended to the current user.
The information recommendation device provided in the embodiment of the invention determines a target user corresponding to a current user, wherein the target user comprises a user whose similarity to the current user meets a preset requirement, respectively obtains first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user, determines an object appearing in the second historical behavior information but not appearing in the first historical behavior information as an object to be recommended, and recommends related information of the object to be recommended to the current user. By adopting the technical scheme, the target user is a user having certain commonality with the current user, and the information which is related to the target user and is not related to the current user is recommended to the current user, so that the technical effect of improving the diversity and novelty of the recommended information while considering the recommendation accuracy can be achieved.
Optionally, the determining the target user corresponding to the current user includes:
respectively calculating the similarity between the current user and each user in a preset user set;
and determining the user with the similarity meeting the preset requirement as a target user.
Optionally, the apparatus may further comprise:
the index measurement module is used for measuring each user in the designated user group according to a preset measurement index before the similarity between the current user and each user in the preset user set is respectively calculated;
and the preset user set generating module is used for generating the preset user set according to the users with the weighing results meeting the preset conditions.
Optionally, the preset metrics include at least one of diversity of shopping behaviors, frequency of shopping behaviors, effective shopping evaluation rate, and whether attack detection passes or not.
Optionally, attack detection is performed in the following manner:
constructing a feature vector of a first user based on shopping evaluation information of the first user;
inputting the feature vector into a preset classifier, and determining whether the first user passes attack detection or not according to an output result of the preset classifier, wherein the preset classifier is generated based on a support vector machine and a K nearest neighbor algorithm.
Optionally, the calculating the similarity between the current user and each user in the preset user set includes:
and respectively calculating the similarity between the current user and each user in the preset user set based on the user attribute and/or the shopping behavior.
Optionally, the first historical behavior information and the second historical behavior information include historical shopping behavior information;
the determining, as an object to be recommended, an object that appears in the second historical behavior information but does not appear in the first historical behavior information includes:
and determining the commodities which appear in the second historical behavior information but do not appear in the first historical behavior information as objects to be recommended.
Embodiments of the present invention also provide a storage medium containing computer-executable instructions, which when executed by a computer processor, perform a method for information recommendation, the method including:
determining target users corresponding to current users, wherein the target users comprise users with similarity to the current users meeting preset requirements;
acquiring first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
determining an object which appears in the second historical behavior information but does not appear in the first historical behavior information as an object to be recommended;
and recommending the relevant information of the object to be recommended to the current user.
Storage medium-any of various types of memory devices or storage devices. The term "storage medium" is intended to include: mounting media such as CD-ROM, floppy disk, or tape devices; computer system memory or random access memory such as DRAM, DDRRAM, SRAM, EDORAM, Lanbas (Rambus) RAM, etc.; non-volatile memory such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. The storage medium may also include other types of memory or combinations thereof. In addition, the storage medium may be located in a first computer system in which the program is executed, or may be located in a different second computer system connected to the first computer system through a network (such as the internet). The second computer system may provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media that may reside in different locations, such as in different computer systems that are connected by a network. The storage medium may store program instructions (e.g., embodied as a computer program) that are executable by one or more processors.
Of course, the storage medium provided by the embodiment of the present invention contains computer-executable instructions, and the computer-executable instructions are not limited to the information recommendation operation described above, and may also perform related operations in the information recommendation method provided by any embodiment of the present invention.
The embodiment of the invention provides computer equipment, and an information recommendation device provided by the embodiment of the invention can be integrated in the computer equipment. Fig. 9 is a block diagram of a computer device according to an embodiment of the present invention. Computer device 900 may include: a memory 901, a processor 902 and a computer program stored on the memory 901 and executable by the processor, wherein the processor 902 executes the computer program to implement the information recommendation method according to the embodiment of the present invention.
According to the computer equipment provided by the embodiment of the invention, the target user is a user having certain commonality with the current user, and the information which is related to the target user and is not related to the current user is recommended to the current user, so that the technical effects of improving the diversity and novelty of the recommended information while considering the recommendation accuracy can be achieved.
The information recommendation device, the storage medium and the computer device provided in the above embodiments may execute the information recommendation method provided in any embodiment of the present invention, and have corresponding functional modules and beneficial effects for executing the method. Technical details that are not described in detail in the above embodiments may be referred to an information recommendation method provided in any embodiment of the present invention.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. An information recommendation method, comprising:
determining target users corresponding to current users, wherein the target users comprise users with similarity to the current users meeting preset requirements;
acquiring first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
determining an object which appears in the second historical behavior information but does not appear in the first historical behavior information as an object to be recommended;
and recommending the relevant information of the object to be recommended to the current user.
2. The method of claim 1, wherein the determining the target user corresponding to the current user comprises:
respectively calculating the similarity between the current user and each user in a preset user set;
and determining the user with the similarity meeting the preset requirement as a target user.
3. The method according to claim 2, further comprising, before the separately calculating the similarity between the current user and each user in the preset user set:
measuring each user in the designated user group according to a preset measurement index;
and generating the preset user set according to the users with the weighing results meeting the preset conditions.
4. The method of claim 3, wherein the predetermined metrics include at least one of a variety of shopping behaviors, a frequency of shopping behaviors, an effective shopping evaluation rate, and whether attack detection passed.
5. The method of claim 4, wherein attack detection is performed by:
constructing a feature vector of a first user based on shopping evaluation information of the first user;
inputting the feature vector into a preset classifier, and determining whether the first user passes attack detection or not according to an output result of the preset classifier, wherein the preset classifier is generated based on a support vector machine and a K nearest neighbor algorithm.
6. The method according to claim 2, wherein the calculating the similarity between the current user and each user in the preset user set comprises:
and respectively calculating the similarity between the current user and each user in the preset user set based on the user attribute and/or the shopping behavior.
7. The method according to any one of claims 1-6, wherein the first historical behavior information and the second historical behavior information comprise historical shopping behavior information;
the determining, as an object to be recommended, an object that appears in the second historical behavior information but does not appear in the first historical behavior information includes:
and determining the commodities which appear in the second historical behavior information but do not appear in the first historical behavior information as objects to be recommended.
8. An information recommendation apparatus, comprising:
the target user determining module is used for determining a target user corresponding to a current user, wherein the target user comprises a user of which the similarity with the current user meets a preset requirement;
the historical behavior information acquisition module is used for acquiring first historical behavior information corresponding to the current user and second historical behavior information corresponding to the target user;
the object to be recommended determining module is used for determining the objects which appear in the second historical behavior information but do not appear in the first historical behavior information as the objects to be recommended;
and the information recommending module is used for recommending the relevant information of the object to be recommended to the current user.
9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1-7 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910180979.8A CN111695023A (en) | 2019-03-11 | 2019-03-11 | Information recommendation method and device, storage medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910180979.8A CN111695023A (en) | 2019-03-11 | 2019-03-11 | Information recommendation method and device, storage medium and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111695023A true CN111695023A (en) | 2020-09-22 |
Family
ID=72474645
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910180979.8A Pending CN111695023A (en) | 2019-03-11 | 2019-03-11 | Information recommendation method and device, storage medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111695023A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112435059A (en) * | 2020-11-24 | 2021-03-02 | 广州富港生活智能科技有限公司 | Method and device for evaluating value of article in real time, electronic equipment and storage medium |
| CN112561571A (en) * | 2020-12-07 | 2021-03-26 | 深圳市思为软件技术有限公司 | House purchasing intention evaluation method and related equipment |
| CN112651804A (en) * | 2020-12-28 | 2021-04-13 | 中国联合网络通信集团有限公司 | Contract machine recommendation method and system |
| CN112907314A (en) * | 2020-12-28 | 2021-06-04 | 桂林旅游学院 | Support Vector Machine (SVM) -based e-commerce recommendation method |
| CN113158037A (en) * | 2021-03-31 | 2021-07-23 | 支付宝(杭州)信息技术有限公司 | Object-oriented information recommendation method and device |
| CN113360756A (en) * | 2021-05-31 | 2021-09-07 | 北京百度网讯科技有限公司 | Resource recommendation method and device, electronic equipment and medium |
| US11869015B1 (en) | 2022-12-09 | 2024-01-09 | Northern Trust Corporation | Computing technologies for benchmarking |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102567900A (en) * | 2011-12-28 | 2012-07-11 | 尚明生 | Method for recommending commodities to customers |
| US20140143233A1 (en) * | 2012-11-20 | 2014-05-22 | Alibaba Group Holding Limited | Method and system for recommending target object information |
| CN105095219A (en) * | 2014-04-23 | 2015-11-25 | 华为技术有限公司 | Weibo recommendation method and terminal |
| CN107507016A (en) * | 2017-06-29 | 2017-12-22 | 北京三快在线科技有限公司 | A kind of information push method and system |
| CN108629665A (en) * | 2018-05-08 | 2018-10-09 | 北京邮电大学 | A kind of individual commodity recommendation method and system |
| CN108876537A (en) * | 2018-06-15 | 2018-11-23 | 重庆知遨科技有限公司 | A kind of mixed recommendation method for on-line mall system |
| CN109087177A (en) * | 2018-08-21 | 2018-12-25 | 北京京东尚科信息技术有限公司 | To the method, apparatus and computer readable storage medium of target user's Recommendations |
| CN109214893A (en) * | 2018-08-31 | 2019-01-15 | 深圳春沐源控股有限公司 | Method of Commodity Recommendation, recommender system and computer installation |
-
2019
- 2019-03-11 CN CN201910180979.8A patent/CN111695023A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102567900A (en) * | 2011-12-28 | 2012-07-11 | 尚明生 | Method for recommending commodities to customers |
| US20140143233A1 (en) * | 2012-11-20 | 2014-05-22 | Alibaba Group Holding Limited | Method and system for recommending target object information |
| CN105095219A (en) * | 2014-04-23 | 2015-11-25 | 华为技术有限公司 | Weibo recommendation method and terminal |
| CN107507016A (en) * | 2017-06-29 | 2017-12-22 | 北京三快在线科技有限公司 | A kind of information push method and system |
| CN108629665A (en) * | 2018-05-08 | 2018-10-09 | 北京邮电大学 | A kind of individual commodity recommendation method and system |
| CN108876537A (en) * | 2018-06-15 | 2018-11-23 | 重庆知遨科技有限公司 | A kind of mixed recommendation method for on-line mall system |
| CN109087177A (en) * | 2018-08-21 | 2018-12-25 | 北京京东尚科信息技术有限公司 | To the method, apparatus and computer readable storage medium of target user's Recommendations |
| CN109214893A (en) * | 2018-08-31 | 2019-01-15 | 深圳春沐源控股有限公司 | Method of Commodity Recommendation, recommender system and computer installation |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112435059A (en) * | 2020-11-24 | 2021-03-02 | 广州富港生活智能科技有限公司 | Method and device for evaluating value of article in real time, electronic equipment and storage medium |
| CN112561571A (en) * | 2020-12-07 | 2021-03-26 | 深圳市思为软件技术有限公司 | House purchasing intention evaluation method and related equipment |
| CN112651804A (en) * | 2020-12-28 | 2021-04-13 | 中国联合网络通信集团有限公司 | Contract machine recommendation method and system |
| CN112907314A (en) * | 2020-12-28 | 2021-06-04 | 桂林旅游学院 | Support Vector Machine (SVM) -based e-commerce recommendation method |
| CN112651804B (en) * | 2020-12-28 | 2023-07-07 | 中国联合网络通信集团有限公司 | Recommendation method and recommendation system of contract machine |
| CN113158037A (en) * | 2021-03-31 | 2021-07-23 | 支付宝(杭州)信息技术有限公司 | Object-oriented information recommendation method and device |
| CN113360756A (en) * | 2021-05-31 | 2021-09-07 | 北京百度网讯科技有限公司 | Resource recommendation method and device, electronic equipment and medium |
| US11869015B1 (en) | 2022-12-09 | 2024-01-09 | Northern Trust Corporation | Computing technologies for benchmarking |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111695023A (en) | Information recommendation method and device, storage medium and equipment | |
| US20200344313A1 (en) | Systems and methods for contextual recommendations | |
| US20180047071A1 (en) | System and methods for aggregating past and predicting future product ratings | |
| US11263222B2 (en) | System for calculating competitive interrelationships in item-pairs | |
| JP6109290B2 (en) | Identify classified misplacements | |
| CN111709812A (en) | E-commerce platform commodity recommendation method and system based on user dynamic classification | |
| Wang et al. | A strategy-oriented operation module for recommender systems in E-commerce | |
| US20190043071A1 (en) | Method and system for selection, filtering or presentation of available sales outlets in a distributed networked computing environment using spatial and geography based metrics | |
| US8825641B2 (en) | Measuring duplication in search results | |
| Li et al. | Economical user-generated content (UGC) marketing for online stores based on a fine-grained joint model of the consumer purchase decision process | |
| CN113077317A (en) | Item recommendation method, device and equipment based on user data and storage medium | |
| Khodabandehlou | Designing an e-commerce recommender system based on collaborative filtering using a data mining approach | |
| Suresh Kumar et al. | A novel weight-optimized LSTM for dynamic pricing solutions in e-commerce platforms based on customer buying behaviour | |
| CN107093122B (en) | Object classification method and device | |
| Huang et al. | Mining the change of customer behavior in dynamic markets | |
| US20150095183A1 (en) | Methods and Apparatus for Generating Recommendations | |
| Patoulia et al. | A comparative study of collaborative filtering in product recommendation | |
| US20190205341A1 (en) | Systems and methods for measuring collected content significance | |
| Aghdaie et al. | Customer-oriented benefit segmentation: an integrated approach | |
| CN113792220A (en) | Target object recommendation method and device, computer equipment and storage medium | |
| CN112381627A (en) | Commodity scoring processing recommendation method and device under child-care knowledge | |
| Nasir et al. | A Survey and Taxonomy of Sequential Recommender Systems for E-commerce Product Recommendation | |
| Najib et al. | A Systematic Literature Review on the Product Ranking Methods | |
| Shukla et al. | Performance optimization of unstructured E-commerce log data for activity and pattern evaluation using web analytics | |
| TEKLEAB | RECOMMENDATION SYTEM ANALYSIS AND EVALUATION |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |