CN105809475A - Commodity recommendation method compatible with O2O applications in internet plus tourism environment - Google Patents

Abstract

The invention discloses a commodity recommendation method compatible with O2O applications in an internet plus tourism environment. The method can be realized through the following steps. First, a tourist chooses a scenic spot to be visited and accesses all available commodities there from the scenic spot inquiring data base. The weight of each commodity is then initialized by the consideration of their click conversion rate and recorded sales. Records of commodities the tourist has browsed, purchased, added to his favorite or disliked will also be checked so as to update the weights of all the commodities. A collaborative filtering recommendation algorithm will be adopted to further update the weights of the commodities. The commodities are then listed in piles according to their weights from high to low and a certain amount of commodities enjoying the highest weights are then chosen as recommended commodities in a temporary list. Then an Apriori algorithm, a basic algorithm of frequent set of items mining method, will be adopted for possible packages of commodities to be purchased. The temporary list with recommended commodities is divided into two halves where the recommended commodities can be all browsed from high weights to low weights and each commodity will be re-listed so that the commodities at the first top half of the temporary list are the final ones to be recommended to the tourist.

Description

The Method of Commodity Recommendation of O2O application is supported under " the Internet+tourism " environment

Technical field

The present invention relates to the particularly computer network travel information service data processing field of the Internet, particularly under " the Internet+tourism " environment, support the Method of Commodity Recommendation of O2O application.

Background technology

In recent years, along with the raising of social progress and people's living standard, tourism is just becoming a kind of pastime become more and more popular.But when people arrive a sight spot, in the face of commodity varied, diversified, visitor is difficult to know the commodity that oneself really needs, and even has a lot of people to be spoofed when selecting commodity.Therefore, the commercial product recommending based on the tourist attractions of mobile app just can help visitor to find the commodity oneself being likely to need, and can also greatly reduce oneself cheated probability with reference to conventional commodity purchasing and evaluation.On the other hand, for sight spot, it is also possible to improve the overall profit of oneself.In sum, research has become a significantly thing based on the Method of Commodity Recommendation of the tourist attractions of mobile app.

The Method of Commodity Recommendation of tourist attractions belongs to typical commending system problem, has a great difference also.For a client, in the face of the article set of enormous amount, how to be quickly found oneself article interested and just become a problem, it is recommended that the appearance of system to solve this problem just.Such as web film adopts film commending system to recommend the film that may like to user, and shopping website adopts commercial product recommending system to be likely to the commodity wanting to buy to lead referral.Commending system is through existing selection course or similarity relationships excavates the object interested that each user is potential, and then carries out personalized recommendation.Be summed up, the difference according to proposed algorithm, it is presently recommended that system can be divided into following a few class: content-based recommendation system, Collaborative Filtering Recommendation System, structure Network Based commending system and mixing commending system.The core concept of content-based recommendation system is respectively user and article to be set up configuration file, buy or browsed content by analyzing, set up or update the configuration file of user, system can compare the similarity of user and product configuration file, and directly recommends the product most like with its configuration file to user.Collaborative Filtering Recommendation System core concept is to utilize the historical information of user calculate the similarity between user and utilize the neighbours higher with targeted customer's similarity to the evaluation of other products to predict that targeted customer to the fancy grade of specific products and recommends.The core concept of the commending system of structure Network Based is to be left out the content character of user and product, and only they is regarded as abstract node, and the information that all algorithms utilize all is hidden among the choice relation of user and product.The core concept of mixing commending system is the pluses and minuses in conjunction with above-mentioned commending system, maximizes favourable factors and minimizes unfavourable ones.

Many Method of Commodity Recommendations for different field are had to have been achieved for some achievements at present.Chinese patent " the recommendation method and system of a kind of product information ", application number: 201010273633.1, Authorization Notice No. 102385601A discloses the recommendation method and system of a kind of product information, the method carries out commercial product recommending only according to this user historical record, it does not have use the hobby data of similar users to recommend.Chinese patent " a kind of e-commerce website Method of Commodity Recommendation based on keyword ", application number: 201210050057.3, Authorization Notice No. 102629257A discloses a kind of e-commerce website Method of Commodity Recommendation based on keyword, the method carries out commercial product recommending only according to this user historical record, the recommendation order of commodity is not carried out data process, also without considering that the data sold between commodity influence each other relation.

But, the commercial product recommending system based on sight spot is understood the place along with visitor arrives due to each recommendable commodity set and be changed, rather than the commodity set only one of which of whole commending system.On the other hand, it is contemplated that between commodity, would be likely to occur the tendency that multiple commodity are bought together, these commodity are put together and is likely to improve the desire to buy of visitor, and the two aspect factor that above-mentioned various commending system need not consider just.Therefore existing commending system can not directly apply to the commercial product recommending based on sight spot.

Summary of the invention

Goal of the invention: present invention is generally directed to the commercial product recommending based on sight spot, make up existing article proposed algorithm deficiency under this is suitable for background, there is provided and under a kind of " the Internet+tourism " environment, support O2O (OnlineToOffline, on line under the line) Method of Commodity Recommendation applied.

In order to solve above-mentioned technical problem, the invention discloses the Method of Commodity Recommendation supporting O2O application under " the Internet+tourism " environment, comprise the following steps:

Step 1, according to the sight spot being about to visit that visitor selects, obtains, by scenery spot query data base, all commodity that this sight spot is commercially available, and carries out recommending weights initialisation according to click conversion ratio and historical sales to each commodity；

Step 2, inquires about the inventory records browsing, buy, collect and not liking that this visitor is conventional, uses content-based recommendation algorithm to update the recommendation weight of all commodity in this sight spot；

Step 3, adopts the Collaborative Filtering Recommendation Algorithm based on user to update commercial product recommending weight, namely first uses k nearest neighbor algorithm, finds the user similar to this visitor, then updates the recommendation weight of all commodity in this sight spot according to the hobby of these users；

All commodity at this sight spot are carried out heapsort according to recommendation weight, select and recommend a number of commodity that weight is the highest to form interim Recommendations list by step 4 from high to low；

Step 5, inquires about the record that the different commodity at this sight spot are bought by same user simultaneously, adopts Apriori Frequent Itemsets Mining Algorithm, calculate the commodity set being likely to be bought together；

Step 6, interim Recommendations list is divided into two parts that quantity is equal, interim Recommendations list is traveled through from high to low by commercial product recommending weight, for each commodity, if its commodity purchased possibly together are not in the first half of interim Recommendations list, then the commodity that this is purchased possibly together are adjusted the first half of interim Recommendations list, finally using the first half of interim Recommendations list as final Recommendations list.

In the present invention, when visitor arrives some specific sight spot, recommendable commodity set is just determined therewith, and in step 1, the set ItemSet of all commodity is expressed as ItemSet={item₁, item₂..., item_i..., item_n, wherein item_iRepresenting i-th commodity, 1≤i≤n, n represents total number of commodity, item in commodity set_i=(itemid_i, class_i, salesvolume_i, clicknum_i, weight_i), wherein itemid_iRepresent id, the class of i-th commodity_iRepresent the classification of i-th commodity, salesvolume_iRepresent i-th commodity sales volume up to now, clicknum_iRepresent the number of visits of i-th commodity, weight_iRepresent the recommendation weight of i-th commodity；Then pass through following steps to carry out recommending weights initialisation:

Step 1-1, calculates the click conversion ratio clickconvrate of i-th commodity by equation below_i:

${\mathrm{clickconvrate}}_i=\frac{{\mathrm{salesvolume}}_i}{{\mathrm{clicknum}}_i},$

Step 1-2, by the equation below recommendation weight weight to i-th commodity_iInitialize:

weight_i=salesvolume_i*clickconvrate_i。

Step 2 comprises the steps:

Step 2-1, inquires about the inventory records set ItemHistorySet={h browsing, buy, collect and not liking of this visitor₁, h₂..., h_a..., h_m, m represents total number of inventory records, wherein h_aRepresenting a article inventory records, 1≤a≤m, each inventory records is expressed as h_a=(class_a, feel_a), wherein class_aRepresent the classification of a commodity, feel_aRepresent this visitor's fancy grade to a commodity, feel_aValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

Step 2-2, travels through each inventory records h in the inventory records set of this visitor_a, all commodity are found out the commodity identical with the merchandise classification of this inventory records in this sight spot, as similar commodity, the then feel according to this inventory records_aField updates the recommendation weight of similar commodity: if feel_aEqual to 1, the recommendation weight weight of these similar commodity_gDeduct 1,1≤g≤Num_a, Num_aRepresent and commodity h_aSimilar commodity amount；If feel_aEqual to 3, the recommendation weight weight of these similar commodity_gPlus 1；If feel_aEqual to 4, the recommendation weight weight of these similar commodity_gPlus 2, thus obtain the commercial product recommending weight after first time renewal.

In step 3, adopt the Collaborative Filtering Recommendation Algorithm second time based on user to update the weight of all commodity, comprise the steps:

Step 3-1, the information userinfo of visitor is expressed as userinfo=(userid, age, gender, class1num, class2num, ...), wherein userid represents visitor id, age represents the age, gender represents sex, class1num represents the quantity purchase of the 1st class commodity, class2num represents the quantity purchase of the 2nd class commodity, below by that analogy, then, adopt k nearest neighbor algorithm, find out k the neighbours closest with this visitor, as k most like visitor, wherein the value of k is set by the user using this Method of Commodity Recommendation, general k value 5 < k < 10；

Step 3-2, inquires about this k the similar visitor historical record to all commodity at this sight spot, and the c log is shown as item_history_c=(class_c, feel_c), wherein class_cRepresenting the classification of c commodity, the upper limit of c is the sum of historical record, feel_cRepresent this visitor's fancy grade to c commodity, feel_cValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

With class in lookup inventory records set ItemHistorySet_cThe commodity that field is equal, as similar commodity, the then feel according to every record_cField updates the recommendation weight of similar commodity: if feel_cEqual to 1, by the weight weight of described commodity_hDeduct 1,1≤h≤Num_c, Num_cRepresent and class_cThe quantity of the commodity that field is equal；If feel_cEqual to 3, by the weight weight of described commodity_hPlus 1；If feel_cEqual to 4, by the weight weight of described commodity_hPlus 2, thus obtain the commercial product recommending weight after second time updates.

In step 4, adopting 2q the commodity that heapsort obtains weight maximum to form orderly interim Recommendations list ItemList, wherein q is the number of consequently recommended commodity, specifically includes following steps:

Step 4-1, initializes empty interim Recommendations list ItemList={}, reconstructs according to the advowson of all commodity at this sight spot and builds a most raft, and the heap top element in most raft is commodity；

Step 4-2, the heap top element every time taking out most raft is saved in interim Recommendations list ItemList afterbody, and last element of heap is put into heap top, readjusts and forms it into new most raft；

Step 4-3, repeats step 4-2, until having 2q heap top element or heap in interim Recommendations list ItemList is sky.

In step 5, if interim Recommendations list ItemList is only less than q element, then direct using the commodity in ItemList as recommendation results, otherwise perform following steps:

Step 5-1, inquires about the record that in interim Recommendations list ItemList, each commodity are bought by user, forms set RecordSet={r₁, r₂..., r_f..., r_r, wherein 1≤f≤r, r is purchaser record bar number, and the f article purchaser record r_f=(userid_f, itemid_f), wherein userid_fRepresent the id of f visitor, itemid_fRepresent the id of the commodity of the f visitor's purchase；

Different records in set RecordSet are carried out cartesian product operation according to userid, form new set NewRecordSet={nr by step 5-2₁, nr₂..., nr_g..., nr_s, wherein 1≤g≤s, s is new purchaser record bar number, and the g purchaser record nr_gFor being expressed as nr_g=(num₁, num₂..., num_b..., num_t), wherein num_bRepresent in interim Recommendations list ItemList, whether the b commodity is bought, and t represents the commodity number in interim Recommendations list ItemList, num_bValue is 1 or 0, num_bRepresent that equal to 1 these commodity are purchased, num_bRepresent that equal to 0 these commodity are not purchased；

Step 5-3, implements Apriori Frequent Itemsets Mining Algorithm to set NewRecordSet and obtains the commodity being likely to simultaneously be bought, and the results set calculated is expressed as RelationSet={re₁, re₂..., re_o..., re_u, wherein 1≤e≤u, u is the bar number of relation, each record re in set_oIt is expressed as: re_o=(num_{O, 1}, num_{O, 2}..., num_{O, u}..., num_{O, t}), wherein num_{O, u}It is calculated by Apriori Frequent Itemsets Mining Algorithm to obtain, if commodity o and commodity u is likely to be bought simultaneously, then corresponding num_{O, u}It is 1, is otherwise 0.

Step 6 comprises the steps:

Step 6-1, travel through interim Recommendations list ItemList, if starting point index=1 to be visited, terminal endj=q to be visited, interim Recommendations list ItemList is divided into two parts that quantity is equal, interim Recommendations list ItemList is traveled through from high to low according to commercial product recommending weight, if these commodity occur on a record in set of relationship RelationSet, then other commodity in this record are adjusted the first half of interim Recommendations list ItemList, if namely other commodity are at the first half of ItemList, its position remains unchanged, otherwise these other commodity are put into forward from endj, and endj is deducted the commodity number put into；

Step 6-2, repeats step 6-1 until index is equal to endj, finally using front q the commodity of interim Recommendations list ItemList as final commercial product recommending list.

Compared with prior art, the invention have the advantages that:

(1) relevant commodity candidate collection can be recommended along with sight spot change.

(2) have employed Apriori algorithm and excavate the commodity that visitor buys possibly together, improve visitor and buy the desire of secondary commodities.

(3) evaluation to commodity " not liking " is introduced so that recommend to avoid some unlikely commodity.

Accompanying drawing explanation

Below in conjunction with the drawings and specific embodiments the present invention being done and further illustrate, the above-mentioned and/or otherwise advantage of the present invention will become apparent.

Fig. 1 is the flow chart of the inventive method.

Detailed description of the invention

Below in conjunction with accompanying drawing, the present invention is illustrated.It it is noted that described embodiment is solely for the purpose of illustration, rather than limitation of the scope of the invention.

The invention discloses the Method of Commodity Recommendation supporting O2O application under " the Internet+tourism " environment, the method flow chart is as it is shown in figure 1, comprise the following steps:

Step 1, according to the sight spot being about to visit that visitor selects, obtains, by scenery spot query data base, all commodity that this sight spot is commercially available, and carries out recommending weights initialisation according to click conversion ratio and historical sales to each commodity；

Step 2, inquires about the inventory records browsing, buy, collect and not liking that this visitor is conventional, uses content-based recommendation algorithm to update the recommendation weight of all commodity in this sight spot；

Step 3, adopts the Collaborative Filtering Recommendation Algorithm based on user to update commercial product recommending weight, namely first uses k nearest neighbor algorithm, finds the user similar to this visitor, then updates the recommendation weight of all commodity in this sight spot according to the hobby of these users；

All commodity at this sight spot are carried out heapsort according to recommendation weight, select and recommend a number of commodity that weight is the highest to form interim Recommendations list by step 4 from high to low；

Step 5, inquires about the record that the different commodity at this sight spot are bought by same user simultaneously, adopts Apriori Frequent Itemsets Mining Algorithm, calculate the commodity set being likely to be bought together；

Step 6, interim Recommendations list is divided into two parts that quantity is equal, interim Recommendations list is traveled through from high to low by commercial product recommending weight, for each commodity, if its commodity purchased possibly together are not in the first half of interim Recommendations list, then the commodity that this is purchased possibly together are adjusted the first half of interim Recommendations list, finally using the first half of interim Recommendations list as final Recommendations list.

In the present invention, when visitor arrives some specific sight spot, recommendable commodity set is just determined therewith, and in step 1, the set ItemSet of all commodity is expressed as ItemSet={item₁, item₂..., item_i..., item_n, wherein item_iRepresenting i-th commodity, 1≤i≤n, n represents total number of commodity, item in commodity set_i=(itemid_i, class_i, salesvolume_i, clicknum_i, weight_i), wherein itemid_iRepresent id, the class of i-th commodity_iRepresent the classification of i-th commodity, salesvolume_iRepresent i-th commodity sales volume up to now, clicknum_iRepresent the number of visits of i-th commodity, weight_iRepresent the recommendation weight of i-th commodity；Then pass through following steps to carry out recommending weights initialisation:

Step 1-1, calculates the click conversion ratio clickconvrate of i-th commodity by equation below_i:

${\mathrm{clickconvrate}}_i=\frac{{\mathrm{salesvolume}}_i}{{\mathrm{clicknum}}_i},$

Step 1-2, by the equation below recommendation weight weight to i-th commodity_iInitialize:

weight_i=salesvolume_i*clickconvrate_i。

Step 2 comprises the steps:

Step 2-1, inquires about the inventory records set ItemHistorySet={h browsing, buy, collect and not liking of this visitor₁, h₂..., h_a..., h_m, m represents total number of inventory records, wherein h_aRepresenting a article inventory records, 1≤a≤m, each inventory records is expressed as h_a=(class_a, feel_a), wherein class_aRepresent the classification of a commodity, feel_aRepresent this visitor's fancy grade to a commodity, feel_aValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

Step 2-2, travels through each inventory records h in the inventory records set of this visitor_a, all commodity are found out the commodity identical with the merchandise classification of this inventory records in this sight spot, as similar commodity, the then feel according to this inventory records_aField updates the recommendation weight of similar commodity: if feel_aEqual to 1, the recommendation weight weight of these similar commodity_gDeduct 1,1≤g≤Num_a, Num_aRepresent and commodity h_aSimilar commodity amount；If feel_aEqual to 3, the recommendation weight weight of these similar commodity_gPlus 1；If feel_aEqual to 4, the recommendation weight weight of these similar commodity_gPlus 2, thus obtain the commercial product recommending weight after first time renewal.

In step 3, adopt the Collaborative Filtering Recommendation Algorithm second time based on user to update the weight of all commodity, comprise the steps:

Step 3-1, the information userinfo of visitor is expressed as userinfo=(userid, age, gender, class1num, class2num, ...), wherein userid represents visitor id, age represents the age, gender represents sex, class1num represents the quantity purchase of the 1st class commodity, class2num represents the quantity purchase of the 2nd class commodity, below by that analogy, then, adopt k nearest neighbor algorithm, find out k the neighbours closest with this visitor, as k most like visitor, wherein the value of k is set by the user using this Method of Commodity Recommendation, general k value 5 < k < 10；

Step 3-2, inquires about this k the similar visitor historical record to all commodity at this sight spot, and the c log is shown as item_history_c=(class_c, feel_c), wherein class_cRepresent the classification of c commodity, feel_cRepresent this visitor's fancy grade to c commodity, feel_cValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

With class in lookup inventory records set ItemHistorySet_cThe commodity that field is equal, as similar commodity, the feel according to every record_cField updates the recommendation weight of similar commodity: if feel_cEqual to 1, by the weight weight of described commodity_hDeduct 1,1≤h≤Num_c, Num_cRepresent and class_cThe quantity of the commodity that field is equal；If feel_cEqual to 3, by the weight weight of described commodity_hPlus 1；If feel_cEqual to 4, by the weight weight of described commodity_hPlus 2, thus obtain the commercial product recommending weight after second time updates.

In step 4, adopting 2q the commodity that heapsort obtains weight maximum to form orderly interim Recommendations list ItemList, wherein q is the number of consequently recommended commodity, specifically includes following steps:

Step 4-1, initializes empty interim Recommendations list ItemList={}, reconstructs according to the advowson of all commodity at this sight spot and builds a most raft, and the heap top element in most raft is commodity；

Step 4-2, the heap top element every time taking out most raft is saved in interim Recommendations list ItemList afterbody, and last element of heap is put into heap top, readjusts and forms it into new most raft；

Step 4-3, repeats step 4-2, until having 2q heap top element or heap in interim Recommendations list ItemList is sky.

In step 5, if interim Recommendations list ItemList is only less than q element, then direct using the commodity in ItemList as recommendation results, otherwise perform following steps:

Step 5-1, inquires about the record that in interim Recommendations list ItemList, each commodity are bought by user, forms set RecordSet={r₁, r2 ..., r_f..., r_r, wherein 1≤f≤r, r is purchaser record bar number, and the f article purchaser record r_f=(userid_f, itemid_f), wherein userid_fRepresent the id of f visitor, itemid_fRepresent the id of the commodity of the f visitor's purchase；

Different records in set RecordSet are carried out cartesian product operation according to userid, form new set NewRecordSet={nr by step 5-2₁, nr₂..., nr_g..., nr_s, wherein 1≤g≤s, s is new purchaser record bar number, and the g purchaser record nr_gFor being expressed as nr_g=(num₁, num₂..., num_b..., num_t), wherein num_bRepresent in interim Recommendations list ItemList, whether the b commodity is bought, and t represents the commodity number in interim Recommendations list ItemList, num_bValue is 1 or 0, num_bRepresent that equal to 1 these commodity are purchased, num_bRepresent that equal to 0 these commodity are not purchased；

Step 5-3, implements Apriori Frequent Itemsets Mining Algorithm to set NewRecordSet and obtains the commodity being likely to simultaneously be bought.

The groundwork of Apriori Frequent Itemsets Mining Algorithm is on a data set, finds out the item collection meeting certain frequently degree.Frequent degree in the present invention refers to that different commodity often frequently occur on the degree of same purchaser record together.The present invention uses this algorithm to carry out in set of computations NewRecordSet the frequent degree between different commodity, obtains, with this, the commodity set that is likely to simultaneously buy.The results set calculated is expressed as RelationSet={re₁, re₂..., re_o..., re_u, wherein 1≤e≤u, u is the bar number of relation, each record re in set_oIt is expressed as: re_o=(num_{O, 1}, num_{O, 2}..., num_{O, u}..., num_o,t), wherein num_o,uCalculated by Apriori Frequent Itemsets Mining Algorithm, if commodity o and commodity u is likely to be bought simultaneously, then corresponding num_o,uIt is 1, is otherwise 0.

Step 6 comprises the steps:

Step 6-1, travel through interim Recommendations list ItemList, if starting point index=1 to be visited, terminal endj=q to be visited, interim Recommendations list ItemList is divided into two parts that quantity is equal, interim Recommendations list ItemList is traveled through from high to low according to commercial product recommending weight, if these commodity occur on a record in set of relationship RelationSet, then other commodity in this record are adjusted the first half of interim Recommendations list ItemList, if namely other commodity are at the first half of ItemList, its position remains unchanged, otherwise these other commodity are put into forward from endj, and endj is deducted the commodity number put into；

Step 6-2, repeats step 6-1 until index is equal to endj, finally using front q the commodity of interim Recommendations list ItemList as final commercial product recommending list.

Embodiment

The present embodiment employs some scene data of A city and tests.The id of visitor B is it is known that set and finally to recommend altogether 12 commodity, i.e. q=12, and it is available that this sight spot one has 100 commodity, ItemSet={item₁, item₂..., item₁₀₀, item_i=(itemid_i, class_i, salesvolume_i, clicknum_i, weight_i)。

First each commodity being calculated and click conversion ratio: if salesvolume is 0, corresponding weight is the minimum positive number of Double type；Otherwise,Then the weight of these commodity is initialized as weight_i=salesvolume_i*clickconvrate_i。

Secondly inquire about this visitor's historical record according to the id of visitor B and constitute set ItemHistorySet={h₁, h₂..., h_a..., h_m, wherein h_aRepresent a article record h_a=(class_a, feel_a).For each h_a, all commodity are found out the commodity equal with the class field of this record in this sight spot, as similar commodity, the feel according to this record_aField updates the weight of similar commodity, if feel_aEqual to 1, the weight of these similar commodity_gDeduct 1, if feel_aEqual to 3, the weight of these similar commodity_gPlus 1, if feel_aEqual to 4, the weight of these similar commodity_gPlus 2.

Then tourist information be expressed as userinfo=(userid, age, gender, class1num, class2num ...), adopt k nearest neighbor algorithm to calculate and most like 50 visitors of visitor B, the id recording them is SimilarUser={id₁, id₂..., id₅₀, to each similar visitor, inquire about this visitor about the historical record of commodity in ItemSet, update the weight of these all commodity further according to the feel field of every inventory records.

Then ItemSet is carried out heapsort and finds out 24 commodity formation items list ItemList that now weight is maximum, and the element of ItemList is also by the high to Low sequence of weight.

Followed by the record that each commodity in inquiry ItemList are bought by certain user, form set RecordSet=RecordSet={r₁, r₂..., r_f..., r_r, and r_f=(userid_f, itemid_f).The userid field of different elements therein is done cartesian product operation, forms new set NewRecordSet={nr₁, nr₂..., nr_g..., nr_s}.Again NewRecordSet is implemented Apriori algorithm and obtain the commodity being likely to simultaneously be bought, obtain set RelationSet={re₁, re₂..., re_o..., re_u, and re_o=(num_o,1,num_o,2,…,num_o,u..., num_{O, t})。

Finally make index=1, endj=12, from the beginning travel through ItemList, each index=index+1, if certain commodity occurs in re_indexIn, but there are other commodity at re_indexIn and do not appear in front 12 commodity, then these commodity are put into forward from endj successively in ItemList, endj deducts the commodity number put into, steps be repeated alternatively until index equal to endj time stop.Then front 12 commodity in ItemList are recommended as recommendation list.

The invention provides the Method of Commodity Recommendation supporting O2O application under " the Internet+tourism " environment; the method and the approach that implement this technical scheme are a lot; the above is only the preferred embodiment of the present invention; should be understood that; for those skilled in the art; under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications also should be regarded as protection scope of the present invention.The all available prior art of each ingredient not clear and definite in the present embodiment is realized.

Claims (7)

1. under " the Internet+tourism " environment, support the Method of Commodity Recommendation of O2O application, it is characterised in that: comprise the following steps:

Step 1, according to the sight spot being about to visit that visitor selects, obtains, by scenery spot query data base, all commodity that this sight spot is commercially available, and carries out recommending weights initialisation according to click conversion ratio and historical sales to each commodity；

Step 2, inquires about the inventory records browsing, buy, collect and not liking that this visitor is conventional, uses content-based recommendation algorithm to update the recommendation weight of all commodity in this sight spot；

Step 3, adopts the Collaborative Filtering Recommendation Algorithm based on user to update commercial product recommending weight, namely first uses k nearest neighbor algorithm, finds the user similar to this visitor, then updates the recommendation weight of all commodity in this sight spot according to the hobby of these users；

All commodity at this sight spot are carried out heapsort according to recommendation weight, select and recommend a number of commodity that weight is the highest to form interim Recommendations list by step 4 from high to low；

Step 5, inquires about the record that the different commodity at this sight spot are bought by same user simultaneously, adopts Apriori Frequent Itemsets Mining Algorithm, calculate the commodity set being likely to be bought together；

Step 6, interim Recommendations list is divided into two parts that quantity is equal, interim Recommendations list is traveled through from high to low by commercial product recommending weight, for each commodity, if its commodity purchased possibly together are not in the first half of interim Recommendations list, then the commodity that this is purchased possibly together are adjusted the first half of interim Recommendations list, finally using the first half of interim Recommendations list as final Recommendations list.

2. support the Method of Commodity Recommendation of O2O application under " the Internet+tourism " environment according to claim 1, it is characterised in that in step 1, the set ItemSet of all commodity is expressed as ItemSet={item₁, item₂..., item_i..., item_n, wherein item_iRepresenting i-th commodity, 1≤i≤n, n represents total number of commodity, item in commodity set_i=(itemid_i, class_i, salesvolume_i, clicknum_i, weight_i), wherein itemid_iRepresent id, the class of i-th commodity_iRepresent the classification of i-th commodity, salesvolume_iRepresent i-th commodity sales volume up to now, clicknum_iRepresent the number of visits of i-th commodity, weight_iRepresent the recommendation weight of i-th commodity；

Carry out as follows recommending weights initialisation:

Step 1-1, calculates the click conversion ratio clickconvrate of i-th commodity by equation below_i:

${\mathrm{clickconvrate}}_i=\frac{{\mathrm{salesvolume}}_i}{{\mathrm{clicknum}}_i},;$

Step 1-2, by the equation below recommendation weight weight to i-th commodity_iInitialize:

weight_i=salesvolume_i*clickconvrate_i。

3. support the Method of Commodity Recommendation of O2O application under " the Internet+tourism " environment according to claim 2, it is characterised in that step 2 comprises the steps:

Step 2-1, inquires about the inventory records set ItemHistorySet={h browsing, buy, collect and not liking of this visitor₁, h₂..., h_a..., h_m, m represents total number of inventory records, wherein h_aRepresenting a article inventory records, 1≤a≤m, each inventory records is expressed as h_a=(class_a, feel_a), wherein classa represents the classification of a commodity, feel_aRepresent this visitor's fancy grade to a commodity, feel_aValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

Step 2-2, travels through each inventory records h in the inventory records set of this visitor_a, all commodity are found out the commodity identical with the merchandise classification of this inventory records in this sight spot, as similar commodity, the then feel according to this inventory records_aField updates the recommendation weight of similar commodity: if feel_aEqual to 1, the recommendation weight weight of these similar commodity_gDeduct 1,1≤g≤Num_a, Num_aRepresent and commodity h_aSimilar commodity amount；If feel_aEqual to 3, the recommendation weight weight of these similar commodity_gPlus 1；If feel_aEqual to 4, the recommendation weight weight of these similar commodity_gPlus 2, thus obtain the commercial product recommending weight after first time renewal.

4. support the Method of Commodity Recommendation of O2O application under " the Internet+tourism " environment according to claim 3, it is characterised in that step 3 comprises the steps:

Step 3-1, the information userinfo of visitor is expressed as:

Userinfo=(userid, age, gender, class1num, class2num ...),

Wherein userid represents that visitor id, age represent that age, gender represent that sex, class1num represent the quantity purchase of the 1st class commodity, and class2num represents the quantity purchase of the 2nd class commodity；Adopt k nearest neighbor algorithm, find out k the neighbours closest with this visitor, as k most like visitor；

Step 3-2, inquires about this k the similar visitor historical record to all commodity at this sight spot, and the c log is shown as item_history_c=(class_c, feel_c), wherein class_cRepresenting the classification of c commodity, the upper limit of c is the sum of historical record, feel_cRepresent this visitor's fancy grade to c commodity, feel_cValue set be that { 1,2,3,4}, correspondence does not like, browses, collects and buys respectively；

With class in lookup inventory records set ItemHistorySet_cThe commodity that field is equal, as similar commodity, the feel according to every record_cField updates the recommendation weight of similar commodity: if feel_cEqual to 1, by the weight weight of described commodity_hDeduct 1,1≤h≤Num_c, Num_cRepresent and class_cThe quantity of the commodity that field is equal；If feel_cEqual to 3, by the weight weight of described commodity_hPlus 1；If feel_cEqual to 4, by the weight weight of described commodity_hPlus 2, thus obtain the commercial product recommending weight after second time updates.

5. under " the Internet+tourism " environment according to claim 4, support the Method of Commodity Recommendation of O2O application, it is characterized in that, in step 4,2q the commodity that heapsort obtains weight maximum are adopted to form orderly interim Recommendations list ItemList, wherein q is the number of consequently recommended commodity, specifically includes following steps:

Step 4-1, initializes empty interim Recommendations list ItemList, reconstructs according to the advowson of all commodity at this sight spot and builds a most raft, and the heap top element in most raft is commodity；

Step 4-2, the heap top element every time taking out most raft is saved in interim Recommendations list ItemList afterbody, and last element of heap is put into heap top, readjusts and forms it into new most raft；

Step 4-3, repeats step 4-2, until having 2q heap top element or heap in interim Recommendations list ItemList is sky.

6. under " the Internet+tourism " environment according to claim 5, support the Method of Commodity Recommendation of O2O application, it is characterized in that, in step 5, if interim Recommendations list ItemList is only less than q element, then directly using the commodity in ItemList as recommendation results, otherwise perform following steps:

Step 5-1, inquires about the record that in interim Recommendations list ItemList, each commodity are bought by user, forms set RecordSet={r₁, r₂..., r_f..., r_r, wherein 1≤f≤r, r is purchaser record bar number, and the f article purchaser record r_f=(userid_f, itemid_f), wherein userid_fRepresent the id of f visitor, itemid_fRepresent the id of the commodity of the f visitor's purchase；

Different records in set RecordSet are carried out cartesian product operation according to userid, form new set NewRecordSet={nr by step 5-2₁, nr₂..., nr_g..., nr_s, wherein 1≤g≤s, s is new purchaser record bar number, and the g purchaser record nr_gFor being expressed as nr_g=(num₁, num₂..., num_b..., num_t), wherein num_bRepresent in interim Recommendations list ItemList, whether the b commodity is bought, and t represents the commodity number in interim Recommendations list ItemList, num_bValue is 1 or 0, num_bRepresent that equal to 1 these commodity are purchased, num_bRepresent that equal to 0 these commodity are not purchased；

Step 5-3, implements Apriori Frequent Itemsets Mining Algorithm to set NewRecordSet and obtains the commodity being likely to simultaneously be bought, and the results set calculated is expressed as RelationSet={re₁, re₂..., re_o..., re_u, wherein 1≤e≤u, u is the bar number of relation, each record re in set_oIt is expressed as:

re_o=(num_{O, 1}, num_{O, 2}..., num_{O, u}..., num_{O, t}),

Wherein num_{O, u}It is calculated by Apriori Frequent Itemsets Mining Algorithm to obtain, if commodity o and commodity u is likely to be bought simultaneously, then corresponding num_{O, u}It is 1, is otherwise 0.

7. support the Method of Commodity Recommendation of O2O application under " the Internet+tourism " environment according to claim 6, it is characterised in that step 6 comprises the steps:

Step 6-1, travel through interim Recommendations list ItemList, if starting point index=1 to be visited, terminal endj=q to be visited, interim Recommendations list ItemList is divided into two parts that quantity is equal, interim Recommendations list ItemList is traveled through from high to low according to commercial product recommending weight, if these commodity occur on a record in set of relationship RelationSet, then other commodity in this record are adjusted the first half of interim Recommendations list ItemList, if namely other commodity are at the first half of ItemList, its position remains unchanged, otherwise these other commodity are put into forward from endj, and endj is deducted the commodity number put into；

Step 6-2, repeats step 6-1 until index is equal to endj, finally using front q the commodity of interim Recommendations list ItemList as final commercial product recommending list.