1. A cold start method of a commodity recommendation system is characterized by comprising the following steps:

step 1, updating a commodity feature set;

and 2, judging the relationship between the new commodity features and the existing commodity feature set.

2. The method of claim 1, wherein step 1 comprises: preparing a feature set V for an existing commodity_di(ii) a For a new commodity entering the market, setting a feature vector V of the new commodity_dThere are two cases as follows:

in the first case, if V_diLength equal to V_dLength, then V_diKeeping the same;

second case, if V_diLength less than V_dLength, then order V_di＝V_d；

In the first case, V_dLength equal to V_diIndicates the new commodity feature vector V_dAll the characteristics in (1) are the characteristic set V of the existing commodity_diThe characteristics already in (1); update V_diThe method comprises the following steps: at V_diIn, and V_dThe consistent characteristic value is 1, otherwise the value is 0;

in the second case, V_dLength greater than V_diIndicates V_dContaining new features, now for feature set V_diUpdating at feature set V_diIncrease of V_dNew features are included and the value of the new features is 0.

3. The method of claim 2, wherein step 2 comprises:

step 2-1, acquiring a commodity set A, wherein the commodity set A comprises commodities of the same type as the new commodities and commodities of related types; circularly traversing each type of goods Td in the goods set A;

step 2-2, each commodity in the commodity class Td is circularly traversed, and V is subjected to_diAnd V_dPerforming bitwise AND operation and then performing AND operation on the matrix R_f,diPerforming a bit-wise OR operation to obtain new data, and adding R_f,diInitialized to all 0 matrix and new data is stored in matrix R_f,diPerforming the following steps;

step 2-3, to the matrix R_f,diThe characteristic vectors of the corresponding commodities are subjected to OR operation, and the calculated numerical value is stored in the vector R_vPerforming the following steps; if R is_vTerminating the loop for a unit vector;

in the commodity set A, traversing the commodities of the same type, and executing according to the steps 2-2 to 2-3; if R is_vAnd (3) expanding the search range instead of the unit vector, continuously searching in the related commodities, and traversing the commodity types with high similarity to the new commodities by giving higher priority to the commodity types with high similarity to the new commodities according to the steps from step 2-2 to step 2-3 until the unit vector is obtained.

Background

In the e-commerce field, the quantity of network commodities is large, people often do not know how to select the commodities facing mass quantities, merchants also suffer from the fact that no method can be used for recommending the most novel commodities meeting the requirements of consumers to the consumers, and a commodity recommendation system is generated under the background. The large e-commerce platforms develop a recommendation system according to historical behavior records of consumers, such as searching, collecting and purchasing, browsing detailed pages, sharing friends and the like, and recommend similar products bought before or products with potential interest to the consumers, so that the requirements of the consumers are met, and the purchase desire is improved. The recommendation system is a main function of assisting consumers to find products meeting requirements from a large amount of information and helping sellers to improve the attention of commodities.

Since the operation behavior of a newly added product in the e-commerce platform is not yet generated, related data which can be analyzed is few, the characteristic value of the product cannot be accurately extracted, the preference degree of a consumer to the product is predicted, which is a typical cold start problem, and the recommendation of a new product is an important but challenging technology. To address this problem, existing merchandise recommendation systems often use collaborative filtering, content-based filtering, and combinations of these techniques. But both of these methods rely on product similarity and previous transaction data by the consumer. The consumer's shopping tastes and needs change over time, and the system is not aware of the consumer's own or purchasing products for others. It cannot fully rely on the consumer's prior transaction data. The content information suggests only similar goods, not exact goods.

Disclosure of Invention

The invention provides a cold start method of a commodity recommendation system, which does not need historical transaction data of consumers. The core idea is to recommend seed consumers for new goods based on market association, match the characteristics of the new goods with the characteristics of existing goods, and determine whether these characteristics are available in the existing goods characteristics set, in this way, the goods will be recommended faster to the consumers, thereby minimizing cold start problems.

The invention comprises the following steps:

step 1, updating a commodity feature set;

and 2, judging the relationship between the new commodity features and the existing commodity feature set.

The step 1 comprises the following steps: preparing a feature set V for an existing commodity_di(ii) a For a new commodity entering the market, setting a feature vector V of the new commodity_dThere are two cases as follows:

in the first case, if V_diLength equal to V_dLength, then V_diKeeping the same;

second case, if V_diLength less than V_dLength, then order V_di＝V_d；

In the first case, V_dLength equal to V_diIndicates the new commodity feature vector V_dAll the characteristics in (1) are the characteristic set V of the existing commodity_diThe characteristics already in (1); update V_diThe method comprises the following steps: at V_diIn, and V_dThe consistent characteristic value is 1, otherwise the value is 0;

in the second case, V_dLength greater than V_diIndicates V_dContaining new features, now for feature set V_diUpdating at feature set V_diIncrease of V_dNew features are included and the value of the new features is 0.

The step 2 comprises the following steps:

step 2-1, acquiring a commodity set A, wherein the commodity set A comprises commodities of the same type as the new commodities and commodities of related types; circularly traversing each type of goods Td in the goods set A;

step 2-2, each commodity in the commodity class Td is circularly traversed, and V is subjected to_diAnd V_dPerforming bitwise AND operation and then performing AND operation on the matrix R_f,diPerforming a bit-wise OR operation to obtain new data, and adding R_f,diInitialized to all 0 matrix and new data is stored in matrix R_f,diPerforming the following steps;

step 2-3, to the matrix R_f,diThe characteristic vectors of the corresponding commodities are subjected to OR operation, and the calculated numerical value is stored in the vector R_vPerforming the following steps; if R is_vTerminating the loop for a unit vector;

in the commodity set A, traversing the commodities of the same type, and executing according to the steps 2-2 to 2-3; if R is_vAnd (3) expanding the search range instead of the unit vector, continuously searching in the related commodities, and traversing the commodity types with high similarity to the new commodities by giving higher priority to the commodity types with high similarity to the new commodities according to the steps from step 2-2 to step 2-3 until the unit vector is obtained.

The invention provides a commodity similarity model based on characteristics. The model learns the similarity between the existing merchandise feature set and the new merchandise feature set. The new merchandise is associated with its category or related categories (i.e., has some common characteristics). For example, a certain mobile phone belongs to a mobile device, but it has some of the same functions as other devices (e.g., tablet, notebook, camera, etc.). These devices are referred to as correlation devices. The recommender system identifies new product features from the existing feature set that were not previously added.

Has the advantages that: the invention provides a cold start method of a commodity recommendation system, which does not need historical transaction data of consumers. The core idea is to recommend seed consumers for new goods based on market association, match the characteristics of the new goods with the characteristics of existing goods, and determine whether these characteristics are available in the existing goods characteristics set, in this way, the goods will be recommended faster to the consumers, thereby minimizing cold start problems.

Drawings

The foregoing and other advantages of the invention will become more apparent from the following detailed description of the invention when taken in conjunction with the accompanying drawings.

FIG. 1 is a diagram of a merchandise feature set update strategy.

Detailed Description

The invention is further explained below with reference to the drawings and the embodiments.

The invention provides a cold start method of a commodity recommendation system, which specifically comprises the following steps: first, data preparation is performed on the commodity recommendation system. Preparing a set of characteristics of an existing commodity, defined as V_di. For a certain commodity, the value of the feature existing in the feature set is 1, and the value of the feature not existing in the feature set is 0. If there is a new good and a new feature appears, then for the new feature, at V_diThe end of the new feature value is 0 for the existing commodity.

Setting a new commodity into the market, the new commodity having new features, the features being the feature set V_diFeatures that are not present, or that are present in other devices (devices of the same or related type). For new articles, there are only two cases, as described below.

In the first case, if V_diLength equal to V_dLength, then V_diKeeping the same;

second case, if V_diLength less than V_dLength, then order:

V_di＝V_d (1)

in the first case, V_dAnd V_diPhase of lengthAgain, this indicates the feature vector V of the new good_dAre in the existing feature set V_diThe features present in (a). At the same time for V_dAt V_diThe value of the characteristic existing in the (1) is 1, otherwise, the value is 0. In the second case, V_dLength greater than V_diThis indicates V_dIncluding some new features. If new features are present, the existing feature set is updated, at V_diThe new features mentioned above are added at the end, the whole representation is seen in fig. 1.

The types of items are discussed above, with one item being the same type of item and the other item being a related type of item. The following method is proposed for this problem of cold start.

In the first step, the characteristics of new commodities are searched for in the same type of commodities. For example, set V_dRelating to a feature vector of a mobile phone, firstly comparing V of the mobile phone_dWith the existing handset feature set V_diThe relationship (2) of (c). To V_diAnd V_dPerforming bitwise AND operation and then performing AND operation on the matrix R_f,diPerforming a bit-wise OR operation to store the data in the matrix R_f,diIn, R_f,diIt needs to be initialized to the all 0 matrix. For matrix R_f,diThe corresponding commodity vectors in (a) are subjected to OR operation (d)₁ or d₂ or d₃ or d₄Etc.). If the unit vector R is obtained_v(the unit vector is a special vector, e.g. [ 1000 … 0 ]]. As defined in matrix theory), the pair R is terminated_f,diPerforming OR operation between the feature vectors of the corresponding commodities;

since this is now at V_diHas been found at V_dAll the features (R) present in_f,diIs a matrix for bit storage).

If the unit vector is not obtained, the search range needs to be expanded, and the search is continued in the feature set of the related commodity. Second step, for and V_dThe commodity type with high similarity is given higher priority. For example, set V_dFor a mobile phone, it is necessary to find the similarity of the related devices (tablet, iPod, notebook, camera, etc.). Repeat likeThe first step process, until the unit vector is obtained. In the worst case, the unit vector may not be found. However, for new goods entering the market, recommendation systems have minimized the cold start problem.

The algorithm flow is as follows:

step 1, circularly traversing each type of goods Td in a goods set A;

step 2, circularly traversing each commodity in the commodity class Td, V_diAnd V_dPerforming bitwise AND operation and then AND R_f,diPerforming bit-wise OR operation, and updating R with the calculated value_f,diA matrix;

step 3, to the matrix R_f,diIs OR-ed between the corresponding vectors in (d)₁ or d₂ or d₃ or d₄Etc.), the calculated values are stored in a vector R_vPerforming the following steps; if R is_vIs a unit vector, the loop is skipped.

Variable definition:

a: a collection of all items (same type of items and related types of items (e.g., cell phone, tablet, camera, laptop, iPod, etc.);

T_d: a type of the commodity;

d_i: specific goods, d_i∈T_d；

V_di: a set of feature vectors for existing goods;

V_d: a feature vector of a new commodity;

R_f,di: storing a data matrix, and initializing the data matrix into a zero matrix;

R_v: and judging the vector.

Finding the unit vector means that all the features of the new commodity are found in the existing commodity feature set. For example, the new product is an applet 12pro mobile phone, and the features (such as endurance, screen resolution, and brand) of the new product are the same as those of the applet 11 and mate 40 in the existing feature set, so that the applet 12pro mobile phone can be recommended to the user who likes the applet 11 and mate 40, and cold-start product recommendation can be completed.

The present invention provides a cold start method of a merchandise recommendation system, and a method and a way for implementing the same are numerous, and the above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, a plurality of modifications and embellishments can be made without departing from the principle of the present invention, and these modifications and embellishments should also be regarded as the protection scope of the present invention. All the components not specified in the present embodiment can be realized by the prior art.