CN110020136B - Object recommendation method and related equipment - Google Patents

Abstract

The application provides an object recommendation method, which is characterized in that under the condition that a user accessing a webpage object is a target type user, the search mode of the user for the webpage object is determined, different search modes can reflect different access intentions of the user for the webpage object, and the target object searched according to the search modes can accord with the intention of the user. Therefore, the object recommendation method provided by the application can be used for recommending objects meeting the intention of target type users with different intentions. In addition, the application also provides related equipment for object recommendation, so as to ensure the application and implementation of the method in practice.

Description

Object recommendation method and related equipment

Technical Field

The application relates to the technical field of internet, in particular to an object recommendation method and related equipment.

Background

The agent purchase is a novel commodity purchase mode, and a purchaser helps an actual customer to purchase commodities and obtains a return through the difference between the purchase and the sale or a certain proportion of commissions. The goods purchased by the purchaser usually come from an e-commerce platform, and in order to promote the sale of more goods, the e-commerce platform can recommend the goods which may be interested to the purchaser according to the purchasing behavior of the purchaser.

The bottom-level technology implementation manner of the above business model is that the electronic commerce platform identifies a purchaser account from a plurality of user accounts for purchasing goods, and recommends goods of the same or similar categories for a purchaser represented by the purchaser account. In the technical scheme, how to determine the interested commodities of the buyer of the type of the buyer is a technical problem to be solved.

Disclosure of Invention

In view of the above, the present application provides an object recommendation method for recommending objects such as related goods for target type users such as buyers.

In order to achieve the purpose, the technical scheme provided by the application is as follows:

in a first aspect, the present application provides an object recommendation method, including:

if the user accessing the webpage object is the target type user, determining a search mode of the user for the webpage object;

searching a target object according to the searching mode;

and sending the relevant information of the target object to the user.

In a second aspect, the present application provides an object recommendation apparatus, including:

the processor is used for determining a search mode of a user for the webpage object if the user accessing the webpage object is a target type user; searching a target object according to the searching mode;

and the communication interface is used for sending the relevant information of the target object to the user.

In a third aspect, the present application provides an object recommendation apparatus, including:

the search mode determining module is used for determining a search mode of a user for the webpage object if the user accessing the webpage object is a target type user;

the target object searching module is used for searching a target object according to the searching mode;

and the object information sending module is used for sending the relevant information of the target object to the user.

According to the technical scheme, the object recommendation method is provided, and under the condition that the user accessing the webpage object is the target type user, the search mode of the user for the webpage object is determined, different search modes correspond to different access intentions of the user for the webpage object, and the target object searched according to the search modes can meet the access requirements of the user. Therefore, the object recommendation method provided by the application can recommend the object meeting the intention of the target type user with different intentions.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a schematic diagram of an architecture of an object recommendation system provided in the present application;

FIG. 2 is a schematic flow chart of training classification models and identifying users using the models provided herein;

fig. 3 is a schematic flowchart of an object recommendation method provided in the present application;

FIG. 4 is a schematic flow chart illustrating an object recommendation method provided in the present application;

fig. 5 is a schematic structural diagram of an object recommendation device provided in the present application;

FIG. 6 is a schematic diagram of another structure of the object recommendation device provided in the present application;

fig. 7 is a hardware framework diagram of an object recommendation device provided in the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In the business model of purchasing, three main parties are involved, namely a purchaser, an actual purchaser and an e-commerce service provider. One specific application scenario of shopping is that naobao (an e-commerce service provider) can issue some commodity information specifically for a rural user (an actual buyer), and village boy (a shopping agent) retrieves and purchases commodities needed or interested by the rural user in the commodity information according to the needs of the rural user.

As shown in fig. 1, the technical architecture for implementing the business model mainly includes two devices, namely, a server and a terminal; the server is a device which is deployed by an e-commerce service provider and contains information of various commodities, the terminal is a device of a purchaser, and the purchaser logs in the server through the terminal and accesses a webpage object, for example, operations such as browsing, purchasing (shopping cart adding) and purchasing of the commodities are performed according to the requirements of an actual purchaser. It should be noted that the server mainly implements the function of electronic commerce, and therefore may be called as an e-commerce server or an e-commerce platform. The e-commerce server may recommend goods to the purchaser in order to sell more goods and also in order to help the purchaser find desired goods more conveniently. As shown in fig. 1, the server needs to determine a recommended object according to the access data and return the recommended object to the terminal.

How the e-commerce server identifies users such as purchasers among a plurality of visiting users and determines goods in which the users are interested is a technical problem to be solved.

At present, commodity recommendation is a commonly used technology of various e-commerce servers, and the technology can collect behavior data of a user on the e-commerce server for a long time, construct a user model according to the behavior data, and depict user preferences, so as to recommend corresponding commodities according to the user model and the user preferences. However, this general recommendation technique is not suitable for the purchasing scenario, mainly because the number of actual purchasing users served by the purchasing users is large in general, and the behavior data of the purchasing users is the sum of the behavior data of a large number of actual purchasing users. For this reason, the following problems arise if the conventional recommendation technique is used.

First, since the products related to the historical behavior data of the purchaser are a collection of products demanded by a large number of actual purchasing users, and the types of the products included therein are complicated and have a large span, it is difficult to select a recommended product from among these various products. Second, the product recommended by the buyer at a certain time is selected according to the historical behavior data of the buyer, however, the actual buyer served by the buyer at the certain time in the certain time of the purchasing behavior is more likely to be different from the historical purchasing behavior, so the historical behavior data does not have reference meaning.

There is another recommendation technique in the prior art, namely public account based recommendation. The public account can be associated with a plurality of individual sub-accounts, a preference model is established for each individual sub-account according to historical behavior data of each individual sub-account, the preference models of all the individual sub-accounts are weighted to form a collective preference model of the public account, and then the same commodity is provided for the individuals according to the collective preference model.

For example, a travel website is provided with a family public account, the family public account can be associated with individual sub-accounts of a plurality of family members, the website can respectively count the film watching preference model of each family member according to the historical data of each individual sub-account, and the film watching preference model of each family member is weighted to generate the whole film watching preference model of the family. And when a plurality of family members have the common film watching requirement, recommending related films for the family according to the integral film watching preference model of the family.

However, the recommendation technology based on the public account is not suitable for a purchasing recommendation scene, and specific inapplicability can be embodied as the following points.

First, the object recommended by the public account is a group, so a collective preference model is calculated. However, in the purchasing scenario, there is no correlation between actual purchasers served by the purchaser, and the history data of other actual purchasers cannot be used as the recommendation basis of some actual purchasers. Secondly, the individual account associated with the public account is known to the e-commerce server, and the e-commerce server can clearly distinguish which historical data belongs to which individual account, so that respective preference models can be calculated according to the historical data belonging to each individual account.

In the prior art, how to recommend commodities to an account is mainly introduced, and no related technical scheme exists at present for how to identify a user account of a purchaser. The existing public account is explicit, that is, if a user registers the public account, the registration platform will definitely mark the account as the public account, and the e-commerce platform receives an access request, and can determine whether the account is used as the public account according to whether the access account has a mark. However, the purchasing behavior is transparent for the e-commerce platform, and the user data of the purchaser, such as a registered account, does not explicitly mark that the user is the purchaser, so that after the e-commerce platform receives an access request, the e-commerce platform cannot determine that the user account is the account of the purchaser directly according to the user account.

In view of this, the present application provides a user identification method. In one application scenario, the e-commerce server may use this method to identify a type of user, a purchaser, among a large number of visiting users. However, it should be noted that the method is not limited to the shopping scenario, and may be applied to any scenario in which one user assists a plurality of users to access an object provided by the e-commerce server. Therefore, the user identified by the user identification method is not limited to a purchaser, which is a specific example of such a user, and the type of user may be a target type user for convenience of description.

The user identification is mainly based on a pre-trained identification model, and the process of the identification model training mainly comprises the following steps A1-A3.

A1: sample user data is obtained.

The sample user data includes positive sample user data and negative sample user data. The positive sample user data refers to user data marked as a target type user, and it should be noted that the user data marked as a positive sample is an explicit target type user.

For example, in a village and village service scene of Taobao, personnel at a village service station register a "village II" account on an e-commerce platform of the village and village service, and use the "village II" account to perform a purchasing service for villagers, so that the registered "village II" accounts are user data of specific target type users, and further the account data of the "village II" accounts can be used as positive sample user data. Of course, the foregoing village service is merely an illustration of a specific service scenario, and should not be a limitation to the application scenario of the present application.

The negative sample user data is some user data randomly obtained among user data that is not labeled as a target type user.

A2: from the sample user data, user features are extracted.

Wherein user features of which aspects are pre-set are extracted.

For example, the number of active days, average daily activity, categories of purchased commodities, etc. of the user in a certain time are preset and extracted. The number of active days of the user in a certain time can be the number of days that the user accesses the object on the E-commerce server in the certain time; the average daily activity may be the length of time that the object on the e-commerce server is accessed on an average daily basis.

More specifically, for example, the user features for extracting the following aspects are preset for the user:

(a) Active days within 3 days, 7 days and 15 days;

(b) The amount of the transaction and the total amount of the transaction within 3 days, 7 days and 15 days;

(c) Categories of goods clicked, purchased, and purchased within 3 days, 7 days, and 15 days;

(d) Daily activity duration of 3 days, 7 days and 15 days;

(e) Becomes the proportion of the search commodity of the purchase commodity in the total search.

The time length and aspect features in the above examples are merely illustrative and the present application is not limited thereto.

It can be understood that the target type user, such as a purchaser, needs to help a large number of users perform the accessing operation (the accessing operation includes logging in the e-commerce server and accessing the object on the e-commerce server), and therefore, the target type user has any one or more of the following characteristics, such as a high activity level, a large number of categories of the accessed object, and a high transaction conversion rate of the goods, compared with the non-target type user. The feature extraction in the above example is set because the features of these aspects can reflect the characteristics of the above target type users.

A3: and training the user characteristics to obtain a recognition model.

The user characteristics extracted from the sample user data may be trained by using an existing classification algorithm, such as a Logistic Regression (LR) algorithm, and the trained model may classify other users to identify whether the other users are target type users. Thus, the model may be referred to as a recognition model or a classification model.

In summary, as shown in fig. 2, the process of obtaining the model mainly includes three steps, i.e., sample collection, feature extraction, and model training. In addition, some features of the feature extraction may be activity of the user, category distribution of the object accessed by the user, purchase of the user, deal of the object accessed by the user, and the like. Of course, in practical applications, the extracted features are not limited thereto.

After the recognition model is obtained through training, when the type of a certain user needs to be recognized, the features of the preset aspects can be extracted from the user data, the extracted user features are input into the recognition model, and the recognition model can determine whether the user is a target type user.

In the case that the user is identified as a target type user, some objects can be further recommended for the target type user. The object may be a product or another target. The way of identifying whether the user is the target type user can be other ways, and is not limited to the use of the training model for identification.

It should be noted that the access of the target type user to the web page object may include various situations. For example, in the shopping application scenario, the shopper may purchase a commodity for the customer in some cases, and may browse the commodity web page randomly in other cases to store experience for the subsequent commodity purchase. In different situations, different types of objects need to be recommended for the user.

The application provides an object recommendation method, and an object recommended by the method can be a commodity or an article with other properties. Referring to fig. 3, a flow of the item recommendation method provided by the present application is shown, and specifically includes the following steps S301 to S303.

S301: and if the user accessing the webpage object is the target type user, determining the search mode of the user on the webpage object.

After receiving an access request of a user to a webpage object, the type of the user can be identified. The specific identification method may identify the type of the user according to the user identification method. If the user belonging to the target type is identified, the search mode of the user for the accessed webpage object can be further determined.

It should be noted that the search pattern may reflect the access intention of the user to the web page object, and the access intention may include, but is not limited to, two types, precise search and random browsing. Taking a purchasing application scene as an example, a precise searching, that is, a purchaser searches or purchases a specific type of commodity, and randomly browses, that is, the purchaser looks at the commodity website approximately, analyzes the characteristics of the current commodity, and makes experience storage for recommending some characteristic commodities for other customers.

In one example, the specific implementation manner of determining the search mode of the user for the webpage object is to obtain an access path of the user for the webpage object; and determining a search mode of the user for the webpage object according to the access path. Specifically, the access path includes webpage modules accessed by the user in the webpage, and may include the duration and the number of times of accessing the webpage modules.

Different webpage modules can express the intention of the user to the webpage object, which webpage modules are contained in the access path is determined, and the number of access times and the access duration of the webpage modules can determine which intention of the user to the webpage object. For example, the web page module may include a home page, a search page, an object detail page, and the like; the homepage indicates that the user intends to browse the webpage object randomly, the search page indicates that the user intends to search the webpage object accurately, and the object detail page also indicates that the user intends to search the webpage object accurately. Therefore, if the user visits the home page for a long time or a large number of times, the intention of the user is random browsing; if the user visits the retrieval page and the object detail page for a long time or a large number of times, the intention of the user is indicated to be accurate searching.

In addition, the object categories accessed by the user can be determined according to the webpage module, and the access intention of the user to the webpage object can also be determined according to whether the object categories have diversity or not. Specifically, if the object categories have diversity, it indicates that the access intention of the user to the web page object is random browsing, and if the object categories do not have diversity, it indicates that the access intention of the user to the web page object is accurate search. The basis of this determination is that if the user is in a random browsing state, the objects that the user accesses are various, and if the user is in an accurate searching state, the objects that the user accesses are usually concentrated in a small number of objects. Therefore, the access intention of the user can be determined according to whether the access objects have diversity.

It should be noted that, different web page modules have different ways of determining the object categories. For example, the web page module is a home page, the home page usually includes a plurality of areas, and different areas correspond to objects of different categories, so that the object categories can be determined according to the areas browsed by the user. For another example, if the web page module is a search page, the object category may be extracted according to a search item input by the user on the search page, for example, if the user searches "2017 a new light and thin down jacket female student money in winter", the extracted object category includes the down jacket. If the web page module is an object detail page, the category to which the object belongs in the object detail page is determined, for example, if the object detail page is a detail page of a jean dress, the category to which the object belongs may be determined to be the dress.

As can be seen from the above description, the characteristics of the access path of the user can reflect the access intention of the user to the webpage object. According to the access intention, the search mode of the user for the webpage object can be directly determined. The search pattern indicates how to search for an object according to rules, and whether the searched object is precise and detailed or is fuzzy and extensive. For example, if the access intention is accurate search, the corresponding search mode is accurate search; if the access intention is random browsing, the corresponding search mode is fuzzy search. For ease of description, the determined seek mode may be referred to as a target seek mode.

S302: and searching the target object according to the searching mode.

The searching mode includes preset object searching rules, and the object searching rules are selected from several objects as target objects. Different search modes recommend objects to users with different degrees of accuracy. It can be understood that if the search mode is accurate search, the object search rule obtains the user intention in the current access, and selects an object meeting the user intention as a target object according to the user intention; if the search mode is fuzzy search, the object search rule determines an object with certain characteristics as a target object according to historical access data.

S303: and sending the relevant information of the target object to the user.

The related information of the target object can be searched from the server, and further the related information of the target object can be contained in a webpage and sent to the user, and the webpage may not be a separate webpage and may be contained in the same webpage with other webpage information. Taking a shopping application scenario as an example, assuming that the determined target object is 5 pieces of one-piece dress, pictures of the 5 pieces of one-piece dress can be displayed in a certain area of a shopping cart webpage.

According to the technical scheme, the object recommendation method is provided, and under the condition that the user accessing the webpage object is the target type user, the method determines the search mode of the user for the webpage object, different search modes can reflect different access intentions of the user for the webpage object, and the target object searched according to the search modes can meet the intentions of the user. Therefore, the object recommendation method provided by the application can be used for recommending objects meeting the intention of target type users with different intentions.

As mentioned above, the access intention of the user to the web page object includes two types, namely, accurate search and random browsing, and the search mode corresponding to the access intention includes two types, namely, accurate search and fuzzy search. How the target object is searched by the two search modes is explained below.

If the search mode is accurate search, the specific implementation manner of searching the target object according to the search mode includes the following steps B1 to B3.

B1: and acquiring access data generated by the user in the process of accessing the webpage object.

The user is a target type user, and the target type user has the characteristics that the accurate searching process of the webpage object is independent every time and has no association relation with the historical accurate searching process. Taking the shopping application scenario as an example, the shopper accesses the webpage to shop for a certain type of goods, and the goods bought this time are not related to the goods bought historically, so the characteristics of the goods bought historically cannot provide reference for the goods bought this time.

Therefore, the access data acquired in this step is the access data generated by the user in the current access process. Specifically, a user's access to a web page object is to continuously generate data of the user's access to the web page object over a period of time. At a certain time point in the access process, access data before the time point in the access process are obtained, and the objects of which categories are accurately searched by a user are determined according to the access data.

Taking a shopping-for-others application scenario as an example, suppose that a user executes the following access operation when accessing website commodities, browses a home page, inputs search items for multiple times, and after obtaining search results, clicks some commodities in the search results, places some commodities in a shopping cart, and purchases some commodities. The user performs each access operation, and the server obtains corresponding access data, such as what the input search term is, what commodities are clicked, what commodities are purchased, and the like. After the server obtains the access data, the following two steps can be executed according to the access data. It should be noted that, as long as the user performs the access operation, the server may obtain corresponding access data, and the server may recommend the target object for the user according to the obtained access data for multiple times at preset time intervals. Generally, the more access data the server has, the more accurate the target object is retrieved, and thus, each subsequent recommended object may be more accurate than the previous recommended object during one web page object access.

If the candidate object satisfying the condition cannot be acquired due to shortage of access data or the like at the time of previous recommendation or at the time of a certain recommendation, an object having some characteristics specified from the history access data may be set as the target object. Some of which may include, but are not limited to, fast moving, low cost, etc.

B2: different types of retrieval sources are obtained from the access data, and different types of alternative objects are retrieved according to the different types of retrieval sources.

The access data can indicate which access operations the user has performed, and the access operations can reflect what the user wants to find, so that various access operations can be extracted from the access data as a retrieval source, the retrieval source is used for retrieving corresponding objects, and the retrieved objects are used as alternative objects.

For example, the search sources include any of the following: the target type operation points to the webpage object, the search word and the position area where the user is located. The three types of search sources and how to obtain the candidate object according to the search sources are described separately.

A first type of search source. In the shopping application scenario, the target type operation may refer to a click operation, a shopping cart adding operation, a purchase operation, and the like, and then the web page object pointed by the target type operation respectively refers to a clicked commodity, a commodity added to a shopping cart, and a purchased commodity. These types of operations performed by the user on these objects indicate that the user is interested in these objects, and the objects pointed to by these operations are most likely the objects that the user wants to find accurately.

In the above case, when retrieving the candidate object, an object having a preset similarity with the web page object pointed by the target type operation may be retrieved as the candidate object. For example, according to the item clicked by the user, the item added to the shopping cart, or the purchased item, items similar to the respective items are searched for as candidate recommended items. In the search, an object having a similarity to a certain object may be calculated using a similarity calculation method. The similarity calculation method may be, but is not limited to, a collaborative filtering method, a content-based similarity calculation method, and the like.

A second type of search source. The search term is the search content input by the user in the search engine, and can also reflect the search intention of the user. Therefore, the search term can be used as a search source, and the candidate object can be further searched by using the search term which is searched by the user.

If the search source includes the input search term, when the candidate object is searched, the search category corresponding to the search term may be determined, and the object belonging to the search category may be searched as the candidate object. Specifically, the search term input by the user may be some objects with detailed characteristics, and in order to find more similar objects to the object searched by the user, the category to which the object belongs may be determined, and the object category is used for searching. A plurality of category items may be included in the category of objects, which may include, but are not limited to, categories, brands, groups of people, styles, and the like. For example, the search term input by the user includes "2017 new thin down jacket female student money in winter," and the search category determined according to the search term is "female _ bosden _ thin down jacket," so that the candidate is searched according to the search category.

A third type of search source. In some cases, the location area where the user is located may reflect what the user wants to find. Taking the shopping application scenario as an example, the location area where the user is located may reflect the location of the customer served by the user, such as a rural shopper, which is usually in a village with the customer served by the shopper. In this case, after obtaining the location area where the user is located, the location area of the customer served by the user can be determined, and further, what the user wants to search for can be determined according to the commodity searched by the user and the relationship between the location area and the commodity.

That is, if the search source includes the location area where the user is located, when searching for the candidate, the object belonging to the search category may be searched for as the candidate object according to the search category preset in the location area.

For example, the location area of the purchaser is in the south, the searched goods are down jackets, and it can be determined that the goods searched for are down jackets with thin money; if the location area of the purchaser is in the north, the retrieved goods are still down jackets, and it can be determined that the goods searched by the purchaser are possibly heavy down jackets. For another example, the location area of the purchaser is at Zhejiang, the retrieved commodity is tea, and as the experience shows that Zhejiang has a preference for dragon well and white tea, the commodity which the purchaser wants to search can be determined to be the dragon well and the white tea; if the location area of the buyer is in Fujian, the searched goods are tea leaves, and the good is known to be more favorable to Tieguanyin according to experience, then the goods which the buyer wants to search for can be determined to be Tieguanyin.

With regard to the above-mentioned several different types of search sources, it should be noted that when a search source is used to search for a candidate, a plurality of search sources may search for the corresponding candidate at the same time.

For example, a term may be combined with an object pointed to by a target type operation. Specifically, after different retrieval categories are obtained according to different retrieval words according to a plurality of retrieval words input by a user, weighting processing is performed on the retrieval categories according to whether the user clicks, purchases or purchases objects under some retrieval categories. For example, if a user clicks, buys, or purchases an object under certain search categories, the weight of the search categories is increased to increase the rank of the search categories. And selecting the retrieval category with higher level weight for searching the alternative object for the retrieval category after the weighting processing. As another example, a search term may be associated with a location area. For explanation and examples, reference may be made to the third type of search source described above.

B3: and determining the candidate object which can accurately correspond to the access intention of the user as the target object.

The alternative objects are searched according to the search source, and different types of alternative objects can be found by different types of search sources. Because different types of search sources have different degrees of accuracy in reflecting the user's intention, the searched candidates have different degrees of accuracy in reflecting the user's access intention.

In one example, the accuracy level set in advance for each type of retrieval source may be obtained; and selecting the target object from the candidate objects according to the sequence of the accuracy level from high to low.

Specifically, different accuracy levels may be set in advance for different types of search sources, for example, the search source includes an object to which a target type operation points, a search term, and a location area where a user is located, and according to descriptions of the three types of search sources, the accuracy degree of the three types of search sources that can reflect the access intention of the user gradually decreases, so that the accuracy levels of the three types of search sources are from high to low.

Therefore, according to the accuracy level of the retrieval source, the candidate objects obtained by the retrieval source are ranked, and the target object is selected from the candidate objects ranked in the front. The manner of selection may be to select a preset number of candidates as target objects.

In another example, the attribute value of the candidate object on the target attribute item may be obtained; and selecting the target object from the candidate objects according to the sorting rule of the attribute values.

Specifically, the candidate may have attribute values such as sales volume, price, deal conversion rate, seller level, and commodity type on the target attribute item. The candidate objects may be sorted according to a sorting rule of a certain attribute value, and some candidate objects sorted before or after are selected as target objects. For example, sorting according to a rule that prices are from low to high, and then selecting the top-ranked candidate as the target object.

The above describes how to recommend a relevant object for a user in case the user wants to find the object accurately. Compared with the existing recommendation mode, the data generated by the user in the access process is used as the recommendation basis, so that the recommended object is more accurate. In addition, in the method, the retrieval sources of different types have different accuracy levels, and the object corresponding to the retrieval source with higher accuracy level is preferentially selected as the recommendation object, so that the accuracy of the recommendation object can be further improved.

In addition to finding such access intents accurately, the access intents of target type users to web page objects may also be arbitrary browsing. For such access intention, how to search for the target object according to the fuzzy search mode corresponding to such access intention is explained.

Specifically, if the search mode is fuzzy search, the following steps C1 to C3 may be specifically included according to the implementation manner of searching for the target object by the search mode.

C1: obtaining target retrieval categories counted based on the historical access data, wherein the target retrieval categories comprise any one or more of the following items: season-based search categories, hot-market search categories, or new popular search categories.

In this case, the user may be interested in knowing information such as popular objects, season-oriented objects or popular objects so as to purchase more highly the client with the intention of freely browsing such access. Therefore, in order to satisfy the user's demand under such access intention, some search categories may be available, which are required to have a season-taking characteristic, a hot-selling characteristic, or a new popularity characteristic. The retrieval category is counted based on the historical access data, which may include historical access data of a plurality of users, that is, the retrieval category is counted from a plurality of user access data.

Regarding season object. There are categories of objects that have a significant seasonal period and such objects may be referred to as seasonal objects, which may be recommended according to a time dimension. Statistics of seasonal commodities are measured by the difference in monthly distribution of sales for each category object. In particular, the monthly mean value and variance of sales volume of various category objects in one year can be counted, and the category objects with large variance have obvious seasonal differences, so that the month with actual sales volume higher than the mean value of the sales volume by a certain threshold can be selected as the hot sales season of the category objects for the category objects.

Regarding popular objects. For the search terms corresponding to each category object, the extraction of the key words can be carried out, and the key word difference in different periods can be compared. When some keywords do not exist in the past statistics or the number of retrieved users is small and the retrieval frequency at the current time point is high, the keywords are considered to have a popular tendency, so that the object having the keywords can be determined as a popular object.

C2: and retrieving the object belonging to the target retrieval category as the candidate object.

Wherein the search category is some of the broad attributes of the object, such as dress. When retrieving the candidate, it is sufficient to retrieve the object belonging to the retrieval category, e.g. the retrieved dress may comprise a close-fitting dress, a loose-fitting dress, etc.

C3: and selecting a target object meeting a preset condition from the candidate objects.

Because the target retrieval categories may be multiple, the alternative objects obtained by different retrieval categories may be sorted according to a certain sort rule, and some objects are selected as target objects after sorting. The sorting method may refer to the method in step B3, and is not described herein.

The above describes how to recommend relevant objects for a user in the case where the user wants to browse freely. The recommended objects are coarse-grained recommendations, but can meet the requirements of users for understanding current fashion trends, season-oriented objects and hot-market objects. The recommended object can help the target type user to grasp the general trend of the object, and the recommendation capability of the target type user for the client is improved.

In summary, as shown in fig. 4, the object recommendation method provided in the present application may include the following parts as a whole. The first part, classification model construction and user type identification, may correspond to the user identification method described above. And a second part, if the user type is determined to belong to the target type, determining the access intention of the user to the webpage object. And if the access intention is accurate search, recommending the objects with fine granularity, and if the access intention is random browsing, recommending the objects with coarse granularity. This section may correspond to the flow shown in fig. 3, steps B1 to B3, and steps C1 to C3.

The following describes a configuration of an object recommendation apparatus provided in the present application. Referring to fig. 5, a structure of an object recommendation device provided in the present application is shown, which specifically includes: a search mode determining module 501, a target object searching module 502 and an object information sending module 503.

A search mode determining module 501, configured to determine, if a user accessing a web page object is a target-type user, a search mode of the user for the web page object;

a target object searching module 502, configured to search a target object according to the search mode;

an object information sending module 503, configured to send the relevant information of the target object to the user.

It should be noted that, when performing a specific function, a module in the object recommendation apparatus may be performed according to a corresponding method in the object recommendation methods, which is not described herein again.

Referring to fig. 6, another structure of the object recommendation apparatus provided in the present application is shown, and the apparatus may further include a user type determination module 504 and a classification model training module 505 based on the apparatus structure shown in fig. 5.

The user type determining module 504 is configured to determine whether a user accessing the web page object is a target type user according to a pre-trained classification model before determining a search mode of the user for the web page object if the user accessing the web page object is the target type user.

A classification model training module 505 for obtaining sample user data; extracting user features from the sample user data; the user characteristics can embody the characteristics of target type users; and training the user features to obtain a recognition model.

The following describes a structure of the object recommendation apparatus provided in the present application. Referring to fig. 7, which shows a structure of an object recommendation device provided in the present application, the structure may specifically include: memory 701, processor 702, communication interface 703, and bus 704.

A memory 701 for storing program instructions and/or data.

The processor 702, by reading the instructions and/or data stored in the memory 701, is configured to perform the following operations: if the user accessing the webpage object is the target type user, determining a search mode of the user for the webpage object; and searching the target object according to the searching mode.

A communication interface 703, configured to send the relevant information of the target object to the user.

A bus 704 for coupling together the various hardware components of the monitoring device that access the data.

In one example, the processor determines a lookup pattern for the web page object by the user, including: the processor is specifically used for obtaining an access path of the user to the webpage object; and determining a search mode of the user for the webpage object according to the access path.

In one example, the search pattern is a precise search or a fuzzy search.

In one example, if the search mode is an exact search, the searching for the target object according to the search mode by the processor includes: the processor is specifically used for acquiring access data generated by the user in the process of accessing the webpage object; obtaining different types of retrieval sources from the access data, and retrieving different types of alternative objects according to the different types of retrieval sources; and determining the candidate object which can accurately correspond to the access intention of the user as the target object.

In one example, the processor determines, as a target object, a candidate object that can exactly correspond to the user's access intention, including: the processor is specifically used for obtaining the precision levels which are preset for various types of retrieval sources; and selecting the target object from the candidate objects according to the sequence of the accuracy level from high to low.

In one example, the processor determines, as a target object, a candidate object that can exactly correspond to the user's access intention, including: the processor is specifically used for obtaining the attribute value of the candidate object on the target attribute item; and selecting a target object from the candidate objects according to the sorting rule of the attribute values.

In one example, the retrieval sources include any of the following: the target type operation points to the webpage object, the search word and the position area where the user is located.

In one example, the processor retrieves different types of candidate objects according to different types of retrieval sources, including: the processor is specifically configured to, if the retrieval source includes a web page object pointed by a target type operation, retrieve an object having a preset similarity with the web page object pointed by the target type operation as an alternative object; if the retrieval source comprises the retrieval word, determining a retrieval category corresponding to the retrieval word, and retrieving an object belonging to the retrieval category as an alternative object; and if the retrieval source comprises a position area where the user is located, retrieving an object belonging to the retrieval category as an alternative object according to the retrieval category preset with the position area.

In one example, if the search mode is fuzzy search, the processor searches for the target object according to the search mode, including: the processor is specifically configured to obtain a target retrieval category counted based on the historical access data, where the target retrieval category includes any one or more of the following: season-based search categories, hot sales search categories or new popular search categories; retrieving objects belonging to the target retrieval category as candidate objects; and selecting a target object meeting preset conditions from the candidate objects.

In one example, the processor is further configured to determine whether the user accessing the web page object is the target type user according to a pre-trained classification model before determining the search pattern of the user for the web page object if the user accessing the web page object is the target type user.

In one example, the processor is further configured to train a classification model; the processor trains a classification model, including: obtaining sample user data; extracting user features from the sample user data; the user characteristics can embody the characteristics of target type users; and training the user features to obtain a recognition model.

It should be noted that, in the present specification, the embodiments are all described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other.

It is further noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the same element.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (17)

1. An object recommendation method, comprising:

if the user accessing the webpage object is the target type user, determining a search mode of the user for the webpage object;

searching a target object according to the searching mode; if the search mode is accurate search, access data generated by the user in the process of accessing the webpage object at this time is acquired; obtaining different types of retrieval sources from the access data, and retrieving different types of alternative objects according to the different types of retrieval sources; determining an alternative object which can accurately correspond to the access intention of the user as a target object; specifically, if the retrieval source includes a web page object pointed by the target type operation, retrieving an object having a preset similarity with the web page object pointed by the target type operation as an alternative object; if the retrieval source comprises the retrieval word, determining a retrieval category corresponding to the retrieval word, and retrieving an object belonging to the retrieval category as an alternative object; if the retrieval source comprises a position area where the user is located, retrieving an object belonging to the retrieval category as an alternative object according to the retrieval category preset with the position area;

and sending the relevant information of the target object to the user.

2. The object recommendation method of claim 1, wherein the determining the search pattern of the user for the webpage object comprises:

obtaining an access path of the user to a webpage object;

and determining a search mode of the user for the webpage object according to the access path.

3. The object recommendation method of claim 2, wherein the search pattern is a precise search or a fuzzy search.

4. The object recommendation method according to claim 1, wherein the determining, as the target object, the candidate object that can exactly correspond to the user access intention comprises:

obtaining the accuracy level preset for each type of retrieval source;

and selecting the target object from the candidate objects according to the sequence of the accuracy level from high to low.

5. The object recommendation method according to claim 1, wherein the determining, as the target object, the candidate object that can exactly correspond to the user access intention comprises:

obtaining the attribute value of the candidate object on the target attribute item;

and selecting a target object from the candidate objects according to the sorting rule of the attribute values.

6. The object recommendation method of claim 1, wherein the search source comprises any of: the target type operation points to the webpage object, the search word and the position area where the user is located.

7. The object recommendation method of claim 3, wherein if the search mode is fuzzy search, searching for the target object according to the search mode comprises:

obtaining target retrieval categories counted based on the historical access data, wherein the target retrieval categories comprise any one or more of the following: season-based search categories, hot sales search categories or new popular search categories;

retrieving objects belonging to the target retrieval category as candidate objects;

and selecting a target object meeting a preset condition from the candidate objects.

8. The object recommendation method of claim 1, wherein before determining the search mode for the web page object by the user if the user accessing the web page object is the target type user, the method further comprises:

and determining whether the user accessing the webpage object is a target type user or not according to a pre-trained classification model.

9. The object recommendation method of claim 8, wherein the classification model is trained by:

obtaining sample user data;

extracting user features from the sample user data; the user characteristics can embody the characteristics of target type users;

and training the user characteristics to obtain a recognition model.

10. An object recommendation device, comprising:

the processor is used for determining a search mode of a user for the webpage object if the user accessing the webpage object is a target type user; searching a target object according to the searching mode;

the communication interface is used for sending the relevant information of the target object to the user;

if the search mode is accurate search, the processor is specifically configured to acquire access data generated by the user in the process of accessing the webpage object this time; obtaining different types of retrieval sources from the access data, and retrieving different types of alternative objects according to the different types of retrieval sources; determining an alternative object which can accurately correspond to the access intention of the user as a target object; the processor is specifically configured to, if the retrieval source includes a web page object pointed by a target type operation, retrieve an object having a preset similarity with the web page object pointed by the target type operation as an alternative object; if the retrieval source comprises a retrieval word, determining a retrieval category corresponding to the retrieval word, and retrieving an object belonging to the retrieval category as an alternative object; and if the retrieval source comprises a position area where the user is located, retrieving an object belonging to the retrieval category as an alternative object according to the retrieval category preset in the position area.

11. The object recommendation device of claim 10, wherein the processor determines a search pattern for the web page object by the user, comprising:

the processor is specifically used for obtaining an access path of the user to the webpage object; and determining a search mode of the user for the webpage object according to the access path.

12. The object recommendation device of claim 10, wherein the processor determines, as the target object, the candidate object that exactly corresponds to the user's access intention, comprising:

the processor is specifically used for obtaining the precision levels which are preset for various types of retrieval sources; and selecting the target object from the candidate objects according to the sequence of the accuracy level from high to low.

13. The object recommendation device of claim 10, wherein the processor determines, as the target object, the candidate object that exactly corresponds to the user's access intention, comprising:

the processor is specifically used for obtaining the attribute value of the candidate object on the target attribute item; and selecting a target object from the candidate objects according to the sorting rule of the attribute values.

14. The object recommending apparatus of claim 10, wherein if the search mode is fuzzy search, the processor searches for the target object according to the search mode, comprising:

a processor, configured to obtain a target search category counted based on historical access data, where the target search category includes any one or more of: season-based search categories, hot sales search categories or new popular search categories; retrieving objects belonging to the target retrieval category as candidate objects; and selecting a target object meeting a preset condition from the candidate objects.

15. The object recommending apparatus according to claim 10,

and the processor is further used for determining whether the user accessing the webpage object is the target type user or not according to a pre-trained classification model before determining the search mode of the user for the webpage object if the user accessing the webpage object is the target type user.

16. The object recommending apparatus according to claim 15,

the processor is also used for training a classification model;

the processor trains a classification model, including:

obtaining sample user data; extracting user features from the sample user data; the user characteristics can embody the characteristics of target type users; and training the user features to obtain a recognition model.

17. An object recommendation device, comprising:

the search mode determining module is used for determining a search mode of a user for the webpage object if the user accessing the webpage object is a target type user;

the target object searching module is used for searching a target object according to the searching mode; the searching mode is accurate searching, and the target object searching module acquires access data generated by the user in the process of accessing the webpage object; obtaining different types of retrieval sources from the access data, and retrieving different types of alternative objects according to the different types of retrieval sources; determining an alternative object which can accurately correspond to the access intention of the user as a target object; specifically, if the retrieval source includes a web page object pointed by the target type operation, the target object searching module retrieves an object having a preset similarity with the web page object pointed by the target type operation as an alternative object; if the retrieval source comprises a retrieval word, a target object searching module determines a retrieval category corresponding to the retrieval word, and retrieves an object belonging to the retrieval category as an alternative object; if the retrieval source comprises a position area where the user is located, the target object searching module retrieves an object belonging to the retrieval category as an alternative object according to the retrieval category preset in the position area;

and the object information sending module is used for sending the relevant information of the target object to the user.

Images