CN102542046A - Book recommendation method based on book contents - Google Patents

Abstract

The invention discloses a book recommendation method which mainly comprises the following steps of: obtaining the content information of books, mainly including book names, authors, content summaries and directories; calculating the similarity among the books based on the attribute information; and recommending a book according to similarity ranking in combination with inventory, sales and promotion data. According to the book recommendation method, the correlation of the books is calculated only by utilizing the similarity of contents among the books, and finally, the most relevant book is recommended.

Description

A kind of book recommendation method based on book content

Technical field

The present invention relates to field of computer technology, refer more particularly to content-based book recommendation method.

Background technology

In recent years, along with the computer and network development of technology, ecommerce has obtained fast development, and especially online book retail development very rapidly.The online book retail not only can bring considerable income to e-commerce venture, the more important thing is to attract popularity, and bring customer traffic, it has become one of core business of large-scale e-commerce venture.

Because internet book store can present the more contents than entity bookstore on the website, find suitable books fast in order to help the client, also buy more books simultaneously in order to attract clients, need to adopt the book recommendation system to generate book recommendation for the client automatically.

At present, most commending system all is to classify to merchandise classification, ranks through collaborative recommendation or similar merchandise sales then and recommends.Its shortcoming has two: the first, and these commending systems are all classified to general merchandise, does not do optimization to book content, can only the way that books are indiscriminately imitated general merchandise be put into different categories and do recommendation, causes recommendation effect not good; The second, its collaborative recommend method that adopts is for forming marketing scale, and the website that possesses the mass selling data is more suitable, but for the books website of just having reached the standard grade, owing to lack transaction data, is difficult to form effective recommendation.In addition, a lot of commending systems does not all reflect stock position at present, and is finally in short supply to the commodity that the user recommends, and obviously can reduce user experience greatly.

Summary of the invention

In view of this, a kind ofly can reflect book content, the book recommendation method that can combine inventory information, sales promotion information is very useful.

In order to address the above problem, the invention provides a kind ofly based on information such as title, author, synopsis, catalogues, and combine the book recommendation method of inventory data and sales promotion information, its technical scheme comprises:

When books of new adding, at first obtain its content information, and these contents are saved as proper vector.Such as, title is saved as proper vector d _t, the author is saved as proper vector d _a, synopsis is saved as proper vector d _s, catalogue is saved as proper vector d _cThese proper vectors will further be handled and extract keyword and weight, and be used for the characteristic parameter of correlation calculations between the books.

Secondly, carry out the extraction of characteristic speech and the weight calculation of characteristic speech respectively to above-mentioned 4 feature vectors, result of calculation is expressed as:

d＝(w ₁，w ₂，...，w _n)

Wherein, w _iThe number of times of representing characteristic speech i to occur also is the weight of characteristic speech i;

Then, utilize the cosine similarity function that merges based on weight to calculate the correlativity between books, computing formula is following:

C(B _x，B _y)＝a×sim(d _t，x，d _t，y)+b×sim(d _a，x，d _a，y)

+c×sim(d _s，x，d _s，y)+d×sim(d _c，x，d _c，y)

Wherein, C is the degree of correlation between the books, B _xB _yRepresent books x and y, sim (d _x, d _y) be the cosine similarity function, a, b, c, d are the weights of different characteristic vector; Merge four feature vectors d of books _t, d _a, d _s, d _cObtain related coefficient C content-based between books.

Next, according to and different books between the size of related coefficient C, choose maximum these books of N of related coefficient, obtain the recommendation list L0 of these books.

At last, rearrangement obtains final recommendation list L to L0 in conjunction with inventory data and sales promotion information.

The present invention can also strengthen recommendation effect through following method:

The above-mentioned cosine similarity function that merges based on weight is used to calculate correlativity between the books, and its weight value is set at a=b=c=d=25%.

Above-mentioned cosine similarity function sim (d _x, d _y) be used for computational item d _xAnd d _yDegree of correlation, its computing formula is:

$\mathrm{sim}(d_x,d_y)=\frac{{\mathrm{\Σ}}_n{w}_{i,x}\·w_{i,y}}{\sqrt{{\mathrm{\Σ}}_n{w_{i,x}}^2\·{\mathrm{\Σ}}_n{w_{i,y}}^2}}$

Above-mentioned content-based book recommendation method; After obtaining recommendation list L0; At first in L0, remove the stock and be 0 books and obtain in stock book recommendation tabulation L1; Secondly according to sales of books amount size L1 is reordered and obtains L2, come the books of sales promotion among the L2 the most according to the order of sequence at last before, obtain final recommendation list L.

Above-mentioned based on book content; And can combine the book recommendation method of inventory information, sales promotion information can deeply excavate the correlativity between the book content; And combine inventory data, sales promotion information to do book recommendation simultaneously based on book content; Compare traditional recommend method, the present invention can access book recommendation more accurately.Particularly under the situation of just reaching the standard grade, lack sales data, the present invention can set up the association between the books fast, provides book recommendation.

Description of drawings

Fig. 1 shows content-based book recommendation flow process;

Fig. 2 shows the flow process that books upgrade recommendation list L0;

Fig. 3 provides the flow process of book recommendation when showing user's browse graph page face.

Embodiment

Below in conjunction with accompanying drawing and embodiment the present invention is done and to describe in further detail:

Fig. 1 shows content-based book recommendation method step, mainly comprises:

Step S100 obtains title, author, summary, directory information, and above-mentioned information is saved as proper vector d.

Step S101, the characteristic speech extracts, characteristic speech weight calculation.Further processing feature vector d, and extract the weight of its characteristic speech and calculated characteristics speech.

Step S102, to doing correlation calculations one to one between the books, the correlation calculations formula is:

C(B _x，B _y)＝a×sim(d _t，x，d _t，y)+b×sim(d _a，x，d _a，y)

+c×sim(d _s，x，d _s，y)+d×sim(d _c，x，d _c，y)

Calculating formula of similarity is:

$\mathrm{sim}(d_x,d_y)=\frac{{\mathrm{\Σ}}_n{w}_{i,x}\·w_{i,y}}{\sqrt{{\mathrm{\Σ}}_n{w_{i,x}}^2\·{{{\mathrm{\Σ}}_{n}w}_{i,y}}^2}}$

Finally obtain the correlativity ranked list of books, choose forward these books of N of correlativity rank, obtain recommendation list L0.

Step S103; Obtaining on the basis of L0, at first in L0, removing the stock and be 0 books and obtain in stock book recommendation tabulation L1, secondly L1 is being reordered and obtain L2 according to sales of books amount size; Before coming the books of sales promotion among the L2 the most according to the order of sequence at last, obtain final recommendation list L.

Step S104, L recommends the user with recommendation list.

The present invention can also be split as two parts with the book recommendation process and implement: first upgrades book recommendation tabulation L0, and with its storage; Second portion is when the user browses books, reads recommendation list L0, combines stock, sale, promotion data to provide book recommendation simultaneously.Do and describe in further detail below in conjunction with accompanying drawing 2,3 couples of the present invention of accompanying drawing:

Fig. 2 shows the flow process that books upgrade recommendation list L0, mainly comprises:

Step S200 judges whether it is the books of putting on the shelf for the first time.

Step S201 is so the books to putting on the shelf for the first time need obtain it and obtain title, author, summary, directory information, and above-mentioned information is saved as proper vector d.

Step S202, the characteristic speech extracts, characteristic speech weight calculation.Further processing feature vector d, and extract the weight of its characteristic speech and calculated characteristics speech.

Step S203 preserves the proper vector d of current books.

Step S204 is so the books to having put on the shelf read the proper vector d that it has been stored.

Step S205 to doing correlation calculations one to one between the books, finally obtains the correlativity ranked list of books, chooses forward these books of N of correlativity rank, obtains recommendation list L0.

Step S206 stores the recommendation list L0 of current books.

Fig. 3 provides the flow process of book recommendation when showing user's browse graph page face, mainly comprise:

Step S300 reads the recommendation list L0 of current books.

Step S301 removes the stock and is 0 books and obtains in stock book recommendation tabulation L1 in L0.

Step S302, according to sales of books amount size L1 being reordered obtains L2.

Step S303, come the books of sales promotion among the L2 the most according to the order of sequence before, obtain final recommendation list L.

Claims (4)

1. a content-based book recommendation method is characterized in that, comprises the steps:

1) obtains title, author, synopsis, the catalogue of books, foregoing is saved as proper vector d, wherein d respectively _tRepresent title, d _aRepresent the author, d _sThe represent content summary, d _cRepresent catalogue;

2) carry out the characteristic speech to proper vector d and extract and weight calculation, result of calculation is expressed as:

d＝(w ₁，w ₂，…，w _n)

Wherein, w _iThe number of times of representing characteristic speech i to occur also is the weight of characteristic speech i;

3) utilize the cosine similarity function that merges based on weight to calculate the correlativity between books, computing formula is following:

C(B _x，B _y)＝a×sim(d _t，x，d _t，y)+b×sim(d _a，x，d _a，y)

+c×sim(d _s，x，d _s，y)+d×sim(d _c，x，d _c，y)

Wherein, C is the degree of correlation between the books, B _xB _yRepresent books x and y, sim (d _x, d _y) be the cosine similarity function, a, b, c, d are the weights of different characteristic vector;

4) according to the relatedness computation result between the books, choose these books of N in proper order from high to low by the degree of correlation, obtain Recommended Books tabulation L0;

5) according to the inventory data of books and sales data, sales promotion information L0 is resequenced, obtain final Recommended Books tabulation L.

2. method according to claim 1 is characterized in that, the value of proper vector weight a, b, c, d is: a=b=c=d=25%.

3. method according to claim 1 and 2 is characterized in that, cosine similarity function sim (d _x, d _y) computing formula be:

$\mathrm{sim}(d_x,d_y)=\frac{{\mathrm{\Σ}}_n{w}_{i,x}\·w_{i,y}}{\sqrt{{\mathrm{\Σ}}_n{w_{i,x}}^2\·{{{\mathrm{\Σ}}_{n}w}_{i,y}}^2}}$

Wherein, d _x, d _yBe respectively the characteristic component d of books x and y, w _{I, x}Be the number of times that characteristic speech i occurs in x, w _{I, y}It is the number of times that characteristic speech i occurs in y.

4. method according to claim 1 is characterized in that, books stock, sale, sales promotion information are following to the method that L0 reorders:

At first in L0, remove the stock and be 0 books and obtain in stock book recommendation tabulation L1, secondly L1 is reordered and obtain L2 according to sales of books amount size, come the books of sales promotion among the L2 the most according to the order of sequence at last before, obtain final recommendation list L.

Images