CN111882381A - Travel recommendation method based on collaborative memory network - Google Patents
Travel recommendation method based on collaborative memory network Download PDFInfo
- Publication number
- CN111882381A CN111882381A CN202010618323.2A CN202010618323A CN111882381A CN 111882381 A CN111882381 A CN 111882381A CN 202010618323 A CN202010618323 A CN 202010618323A CN 111882381 A CN111882381 A CN 111882381A
- Authority
- CN
- China
- Prior art keywords
- user
- spot
- vector
- sight
- scenery
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 230000007246 mechanism Effects 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 131
- 239000011159 matrix material Substances 0.000 claims description 90
- 238000010606 normalization Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 241000256856 Vespidae Species 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011423 initialization method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/14—Travel agencies
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Tourism & Hospitality (AREA)
- Finance (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a travel recommendation method based on a collaborative memory network, which comprises the steps of firstly using a neural network to vectorize a user and a scenic spot, then introducing a memory network, adopting an attention mechanism method in the memory network, and finally combining global structure information of a hidden factor model with local neighborhood structure information of the memory network when predicting the score of the user on the scenic spot. The invention simultaneously considers the local neighborhood structure information of the user and the local neighborhood structure information of the scenic spots and fuses with the collaborative filtering method of the hidden factor model, thereby realizing the tourism recommendation method based on the collaborative memory network and achieving the goal of high accuracy and high personalized tourism recommendation.
Description
Technical Field
The invention relates to the technical field of intelligent recommendation, in particular to a travel recommendation method based on a collaborative memory network.
Background
With the development of society and the improvement of living standard of people, more and more people choose to travel outside, due to the vigorous development of the tourism industry and the rapid popularization of the internet technology in recent years, the tourism information of some mainstream tourism information service platforms is overloaded at present, the personalized recommendation demand of people for tourist attractions becomes greater and greater, and the recommendation of suitable scenic spots is very important when people travel outside.
The traditional travel recommendation method based on deep learning mainly utilizes a collaborative filtering method based on a hidden factor model to construct a deep learning framework, and thus recommendation is carried out. The recommendation methods only consider the global structure information of the hidden factor model, but not consider the local neighborhood structure information, which causes the problems of low accuracy and insufficient individuation of the scenic spot recommendation result.
Disclosure of Invention
The invention aims to solve the problems of low accuracy and insufficient individuation of scenic spot recommendation results in the conventional deep learning-based tourism recommendation method, and provides a tourism recommendation method based on a collaborative memory network.
In order to solve the problems, the invention is realized by the following technical scheme:
a travel recommendation method based on a collaborative memory network comprises the following steps:
step 1, collecting travel notes of users from a travel website by using a crawler tool, extracting scenic spots visited by each user from the travel notes, and numbering all the users and all the scenic spots respectively;
step 2, taking all scenic spots visited by each user as a set respectively, and constructing a scenic spot neighborhood set related to the user; meanwhile, all users visiting each scenic spot are respectively used as a set to construct a user neighborhood set about the scenic spot;
step 3, mapping all users and all scenic spots to two feature vector spaces respectively by utilizing a neural network, wherein one feature vector space is a user vector matrix and a scenic spot vector matrix, and the other feature vector space is a user external memory matrix and a scenic spot external memory matrix;
step 4, for each user u, calculating the similarity q between the user u and each user v in the user neighborhood set related to the scenic spot i based on the user vector matrix and the scenic spot vector matrix obtained in the step 3uivAnd the similarity q between the user u and each user v in the user neighborhood setuivPerforming normalization processingThen obtaining the weight coefficient p of the user uuiv(ii) a Wherein:
step 5, weighting coefficient p of user uuivAs weights in the attention mechanism, and carrying out weighted summation on user feature vectors of each user v in the user neighborhood set of the sight spot i in the user external memory matrix to obtain neighborhood structure information o of the user uui；
Step 6, for each sight spot i, calculating the similarity q between each sight spot j in the sight spot neighborhood set of the user u and the sight spot i based on the user vector matrix and the sight spot vector matrix obtained in the step 3iujAnd the similarity q of the scenic spot i and each scenic spot j in the scenic spot neighborhood setiujObtaining the weight coefficient p of the sight spot i after normalization processingiuj(ii) a Wherein:
step 7, weighting coefficient p of the scenic spot iiujAs the weight in the attention mechanism, weighting and summing the sight spot feature vectors of each sight spot j in the sight spot field set of the user u in the sight spot external memory matrix to obtain the neighborhood structure information o about the sight spot iiu；
Step 8, performing element product on the user feature vector of the user u in the user vector matrix and the scenery spot feature vector of the scenery spot i in the scenery spot vector matrix to obtain the global structure information of the user u on the scenery spot i;
step 9, obtaining the neighborhood structure information o about the user u obtained in the step 5uiAnd 7, obtaining the neighborhood structure information o about the sight spot iiuSplicing the global structure information of the scenic spot i by the user u obtained in the step 8, and inputting the spliced vector into the multilayer neural network to obtain the final score of the user u on the scenic spot i;
step 10, when a scenic spot needs to be recommended for a certain target user, sequencing the final scores of all the scenic spots of the target user, and selecting K scenic spots with the Top ranks, namely, the Top-K scenic spot recommendation of the target user is obtained;
above, muUser feature vector, m, representing user u in user vector matrixvRepresenting the user characteristic vector of the user v in the user vector matrix; e.g. of the typeiThe sight feature vector representing the sight i in the sight vector matrix, ejExpressing the sight spot characteristic vector of the sight spot j in the sight spot vector matrix; n (i) represents a user neighborhood set about the scenery i formed by all users who visit the scenery i, and S (u) represents a scenery neighborhood set about the user u formed by all the scenery visited by the user u; u is 1,2, …, n, n represents the number of users; i is 1,2, …, m, m represents the number of scenic spots; wherein K is the number of the set recommended scenic spots.
In the step 3, the dimensions of the user vector matrix and the user external memory matrix are both nxd, and the dimensions of the scenery spot vector matrix and the scenery spot external memory matrix are both mxd, where n represents the number of users, m represents the number of scenery spots, and d represents the dimensions of the matrices.
In the step 3, the method further includes the following steps: and (4) performing pre-training updating on the user vector matrix and the sight spot vector matrix by using a generalized matrix decomposition method.
In the above step 5, neighborhood structure information o about user uuiComprises the following steps:
in the formula, puivWeight coefficient representing user u, cvRepresenting a user feature vector of each user v in a user external memory matrix in a user neighborhood set about the sight point i, n (i) representing a user neighborhood set formed by all users visiting the sight point i, u being 1,2, …, n, n representing the number of the users; i is 1,2, …, m, m indicates the number of sights.
In the above step 7, the neighborhood structure information o of the sight spot iiuComprises the following steps:
in the formula, piujWeight coefficient, y, representing the sight ijS (u) represents a scenery feature vector in a scenery external memory matrix of each scenery j in a scenery neighborhood set of the user u, wherein u is 1,2, …, n, and n represents the number of users; i is 1,2, …, m, m indicates the number of sights.
According to the invention, by utilizing a collaborative filtering method based on neighborhood or memory, the local neighborhood structure information of the user and the local neighborhood structure information of the scenic spots are considered at the same time, and are fused with the collaborative filtering method of the hidden factor model, the travel recommendation method based on the collaborative memory network is realized, so that the goals of high accuracy and high personalized travel recommendation are achieved.
Compared with the prior art, the invention has the following characteristics:
1. the invention uses the neural network to vectorize the user and the scenic spots, namely, the user and the scenic spots are mapped into a low-dimensional vector space and expressed by vectors. The method expresses natural language as simple vector, not only maintains the meaning of original data, but also greatly simplifies calculation, so that the expression of the user and the scenic spot is more accurately and reasonably combined with travel recommendation;
2. the memory network is introduced, and the attention mechanism method is adopted in the memory network, so that the similarity between the user and the user in the neighborhood of the user is considered, the similarity between the scenic spot and the scenic spot in the neighborhood of the scenic spot is also considered, and the local neighborhood structure information comprising the user neighborhood and the scenic spot neighborhood is obtained, so that the individuation and the accuracy of travel recommendation can be enhanced;
3. when the score of the user on the scenic spot is predicted, the global structure information of the hidden factor model is combined with the local neighborhood structure information of the memory network, and the global and local information is fully considered, so that the effectiveness of travel recommendation is ensured.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to specific examples.
A travel recommendation method based on a collaborative memory network specifically comprises the following steps:
step 1, collecting travel notes of users from a travel website by using a crawler tool, extracting scenic spots visited by each user from the travel notes, and numbering all the users and all the scenic spots respectively.
The method comprises the steps of collecting travel notes of users on travel websites such as journey taking websites and hornet cells by using the existing web crawler means, extracting scenic spots which are visited by each user according to semi-structured data of scenic spot sequences visited by the users contained in the travel notes of the users, and numbering all the collected users and all the scenic spots respectively to achieve the purpose of data preprocessing.
For the scenic spots visited by the user in the tourist note, all the scenic spots mentioned in the tourist note can be reserved, or the scenic spots can be properly removed according to the user score, that is, the scenic spots with higher scores in the scenic spots visited by the user are reserved, for example, when the extracted comprehensive score of the scenic spots is 2.8, 3.5, 4.5, or the like, that is, five scores are divided, the scenic spots with the score of more than or equal to 3 are reserved.
Because the original travel note data can not be directly used for subsequent calculation, all collected users and scenic spots need to be uniformly numbered and represented by unique ID values, so that the subsequent further processing is facilitated, and the anonymization processing can be performed on the user information. For example: the ID of the first user "03202001" is set to 0, and the ID of the second user "03202002" is set to 1; the ID value of the seven-star scenic spot of the first scenic spot is set to be 0, and the ID of the Xiangshan scenic spot of the second scenic spot is set to be 1; and by analogy, the ID values of the subsequent data are added by 1 one by one.
And 2, respectively taking all the scenic spots visited by each user as a set, and constructing a scenic spot neighborhood set related to the user. And respectively taking all the users visiting each sight spot as a set to construct a user neighborhood set about the sight spot.
All sights visited by the same user are represented by the form of a set < P1, P2, … Pn >, which constitutes a set of sight neighborhoods for the user, wherein P1, P2, … Pn represent the IDs of sights visited by the same user. For example: the user U1 navigates through the sights P1, P2, … … Pn, and the set of sights neighbors for the user U1 is < P1, P2, … Pn >.
All users who have accessed the same attraction are represented by the set < U1, U2, … Un >, constituting a set of user neighborhoods for that attraction, where U1, U2, … Un represent the IDs of users who have accessed the same attraction. For example: users U1, U2, … … Un all visited sight P1, and the user neighborhood set for sight P1 is < U1, U2, … Un >.
And 3, mapping all the users and the scenic spots to two feature vector spaces respectively by utilizing a neural network, wherein one feature vector space comprises a user vector matrix and a scenic spot vector matrix, and the other feature vector space comprises a user external memory matrix and a scenic spot external memory matrix. The dimension of the user vector matrix and the dimension of the user external memory matrix are n × d, the dimension of the scenery spot vector matrix and the dimension of the scenery spot external memory matrix are m × d, n, m respectively represent the number of the user and the scenery spot, d represents the dimension of a given matrix, and in this embodiment, d is 50.
The two parameters of the number n of the users and the number m of the scenic spots are input into a neural network module, and the neural network module outputs a vector matrix with dimension of n x d and related to the users and a vector matrix with dimension of m x d and related to the scenic spots. Since each value in the 2 vector matrices is generated by a self-set initialization method (for example, a normal distribution, a number of 0-1 is randomly generated, and the number is random), inputting the parameters n and m into the neural network twice will obtain different results, first obtaining "an nxd user vector matrix and an mxd scenery spot vector matrix", and second obtaining "an nxd user external memory matrix and an mxd scenery spot external memory matrix".
In addition, in order to make the similarity between the vector of the user and the vector of the sight visited by the user higher than the similarity between the vector of the sight not visited by the user, where the similarity is measured by means of dot product, in this embodiment, a generalized matrix decomposition method (a deep learning model) is further used for pre-training and updating the obtained user vector matrix and sight vector matrix.
And 4, calculating the similarity between the user vector in the user vector matrix and other user vectors in the user neighborhood set as a weight coefficient, and then constructing a user memory network according to the user vector matrix, the corresponding weight coefficient and the user external memory matrix to obtain neighborhood structure information about the user.
And selecting a user with the serial number u from the user set with the uniform serial number, selecting a scenic spot with the serial number i from the scenic spot set with the uniform serial number, and taking the user u and the scenic spot i as the user and the scenic spot constructed by the current memory network.
Firstly, in a user vector matrix, carrying out point multiplication on a vector corresponding to a user u and each user vector in a user neighborhood set of the user u about a scenery i, and then adding the result of the point multiplication to the point multiplication of the vector in the scenery vector matrix corresponding to the scenery i and the user vector in the user neighborhood set to obtain the similarity q between the user u and each user in the user neighborhood set of the user uuiv：
In the formula, quivRepresenting the similarity before normalization, m, of user u and user v in a user neighborhood set for sight iu，mvRespectively representing the feature vectors, e, corresponding to the users u and v in the user vector matrixiRepresenting feature vectors corresponding to the scenery points i in the scenery point vector matrix, and N (i) representing a user neighborhood set formed by all users who visit the scenery points i;
then, the obtained similarity q between the user u and each user in the user neighborhood set is useduivNormalization is carried out, and the similarity after normalization is taken as a weight coefficient p of the user uuiv：
In the formula, exp (x) is an exponential function with e as base and x as index.
Finally, the weight coefficient p of the user u is calculateduivAs weights in the attention mechanism, carrying out weighted summation on feature vectors of each user v in the user neighborhood set of the sight point i in the user external memory matrix, representing the user u through the user vectors in the user neighborhood set of the user u in the user external memory matrix, and thus obtaining a vector representation o containing user neighborhood informationuiI.e. neighborhood structure information about user u. Wherein:
in the formula, puivWeight coefficient representing user u, cvRepresenting the feature vectors of each user v in the user's neighborhood set for sight i in the user's external memory matrix.
And 5, calculating the similarity between the scenery spot vector in the scenery spot vector matrix and other scenery spot vectors in the scenery spot neighborhood set as a weight coefficient, and then constructing a scenery spot memory network according to the scenery spot vector matrix, the corresponding weight coefficient and the scenery spot external memory matrix to obtain neighborhood structure information about the scenery spot.
And selecting a user with the serial number u from the user set with the uniform serial number, selecting a scenic spot with the serial number i from the scenic spot set with the uniform serial number, and taking the user u and the scenic spot i as the user and the scenic spot constructed by the current memory network.
Firstly, in a scenery spot vector matrix, carrying out point multiplication on a vector corresponding to a scenery i and each scenery vector in a scenery spot neighborhood set related to a user u, and then adding the result of the point multiplication to the point multiplication of the vector in the user vector matrix corresponding to the user u and the scenery vector in the scenery spot neighborhood set to obtain the similarity q between the scenery i and each scenery in the scenery spot neighborhood set of the scenery iiuj：
In the formula, qiujRepresenting the similarity of sight i and sight j in a sight neighborhood set about user u before normalization, ei，ejRespectively representing feature vectors m corresponding to the sight points i and j in the sight point vector matrixuS (u) represents a feature vector corresponding to the user u in the user vector matrix, and represents a scenic spot neighborhood set formed by all scenic spots visited by the user u;
then, the similarity q between the sight spot i and each sight spot in the sight spot neighborhood setiujNormalization is carried out, and the normalized similarity is used as a weight coefficient p of the sight spot iiuj：
In the formula, exp (x) is an exponential function with e as base and x as index.
Finally, the weight coefficient p of the sight spot i is calculatediujAs a weight in the attention mechanism, and performing weighted summation on feature vectors of each sight spot j in the sight spot domain set of the user u in the sight spot external memory matrix, that is, expressing the sight spot i by the sight spot vector in the sight spot neighborhood set of the sight spot i in the sight spot external memory matrix, thereby obtaining a vector expression o containing sight spot neighborhood informationiuI.e., neighborhood structure information about the sight i. Wherein:
in the formula, piujWeight coefficient, y, representing the sight ijRepresenting the feature vectors in the sight external memory matrix for each sight j in the sight neighborhood set for user u.
Step 6, performing element product on the feature vector corresponding to the user u in the user vector matrix and the feature vector corresponding to the scenic spot i in the scenic spot vector matrix to obtain vector representation of preference information of the user to the scenic spot, namely the global structure information of the user u of the hidden factor model to the scenic spot i;
step 7, representing the global structure information of the hidden factor model and the vector containing the user neighborhood information (neighborhood structure information o about the user u) in a nonlinear modeui) And a vector representation containing sight neighborhood information (neighborhood structure information o about sight i)iu) And (6) splicing.
For example, if the vector dimensions of the global structure information of the hidden factor model, the vector representation including the user neighborhood information, and the vector representation including the scenery spot neighborhood information are all 100 dimensions, the spliced vector dimension is 300 dimensions.
And 8, inputting the spliced vector into a multilayer neural network to obtain the final score of the user on the scenic spot, wherein the smaller the score value is, the smaller the possibility that the user visits the scenic spot is, and the larger the score value is, the more possible the user visits the scenic spot is. The specific formula is as follows:
rui＝vTφ(U(mu⊙ei)+Woui+Woiu+b) (7)
wherein r isuiThe scores indicating the user's sight spot, U, W, v and b, are parameters that need to be learned in the neural network, indicate the product of elements, i.e. the multiplication of corresponding elements of two vectors, resulting in a new vector with the same dimension as the original vector, and phi (x) indicates the nonlinear activation function.
And 9, calculating the score values of the target user on all the scenic spots through the series of methods, sequencing the obtained scores, and selecting K scenic spots with the Top ranks to obtain the Top-K tourist attraction recommendation of the target user.
The collaborative filtering method based on the hidden factor model and the collaborative filtering method based on the neighborhood or the memory are fused, and the advantages of the global structure information of the hidden factor model and the local structure information of the neighborhood or the memory can be utilized, so that the purpose of personalized travel recommendation is achieved.
It should be noted that, although the above-mentioned embodiments of the present invention are illustrative, the present invention is not limited thereto, and thus the present invention is not limited to the above-mentioned embodiments. Other embodiments, which can be made by those skilled in the art in light of the teachings of the present invention, are considered to be within the scope of the present invention without departing from its principles.
Claims (5)
1. A travel recommendation method based on a collaborative memory network is characterized by comprising the following steps:
step 1, collecting travel notes of users from a travel website by using a crawler tool, extracting scenic spots visited by each user from the travel notes, and numbering all the users and all the scenic spots respectively;
step 2, taking all scenic spots visited by each user as a set respectively, and constructing a scenic spot neighborhood set related to the user; meanwhile, all users visiting each scenic spot are respectively used as a set to construct a user neighborhood set about the scenic spot;
step 3, mapping all users and all scenic spots to two feature vector spaces respectively by utilizing a neural network, wherein one feature vector space is a user vector matrix and a scenic spot vector matrix, and the other feature vector space is a user external memory matrix and a scenic spot external memory matrix;
step 4, for each user u, calculating the similarity q between the user u and each user v in the user neighborhood set related to the scenic spot i based on the user vector matrix and the scenic spot vector matrix obtained in the step 3uivAnd the similarity q between the user u and each user v in the user neighborhood setuivObtaining the weight coefficient p of the user u after normalization processinguiv(ii) a Wherein:
step 5, weighting coefficient p of user uuivAs weights in the attention mechanism, and carrying out weighted summation on user feature vectors of each user v in the user neighborhood set of the sight spot i in the user external memory matrix to obtain neighborhood structure information o of the user uui；
Step 6, aiming at each sceneAnd (3) calculating the similarity q between the scenery i and each scenery j in the scenery neighborhood set related to the user u based on the user vector matrix and the scenery vector matrix obtained in the step (3)iujAnd the similarity q of the scenic spot i and each scenic spot j in the scenic spot neighborhood setiujObtaining the weight coefficient p of the sight spot i after normalization processingiuj(ii) a Wherein:
step 7, weighting coefficient p of the scenic spot iiujAs the weight in the attention mechanism, weighting and summing the sight spot feature vectors of each sight spot j in the sight spot field set of the user u in the sight spot external memory matrix to obtain the neighborhood structure information o about the sight spot iiu；
Step 8, performing element product on the user feature vector of the user u in the user vector matrix and the scenery spot feature vector of the scenery spot i in the scenery spot vector matrix to obtain the global structure information of the user u on the scenery spot i;
step 9, obtaining the neighborhood structure information o about the user u obtained in the step 5uiAnd 7, obtaining the neighborhood structure information o about the sight spot iiuSplicing the global structure information of the scenic spot i by the user u obtained in the step 8, and inputting the spliced vector into the multilayer neural network to obtain the final score of the user u on the scenic spot i;
step 10, when a scenic spot needs to be recommended for a certain target user, sequencing the final scores of all the scenic spots of the target user, and selecting K scenic spots with the Top ranks, namely, the Top-K scenic spot recommendation of the target user is obtained;
above, muUser feature vector, m, representing user u in user vector matrixvRepresenting the user characteristic vector of the user v in the user vector matrix; e.g. of the typeiThe sight feature vector representing the sight i in the sight vector matrix, ejExpressing the sight spot characteristic vector of the sight spot j in the sight spot vector matrix; v ∈ N (i), N (i) indicating the sights formed by all users who have visited sight ii's user neighborhood set; j belongs to S (u), and S (u) represents a scenery neighborhood set which is formed by all sights visited by the user u and is about the user u; u is 1,2, …, n, n represents the number of users; i is 1,2, …, m, m represents the number of scenic spots; wherein K is the number of the set recommended scenic spots.
2. The method as claimed in claim 1, wherein in step 3, the dimensions of the user vector matrix and the user external memory matrix are both nxd, and the dimensions of the scenery spot vector matrix and the scenery spot external memory matrix are both mxd, wherein n represents the number of users, m represents the number of scenery spots, and d represents the dimensions of the matrices.
3. The travel recommendation method based on the collaborative memory network as claimed in claim 1 or 2, wherein the step 3 further comprises the following steps: and (4) performing pre-training updating on the user vector matrix and the sight spot vector matrix by using a generalized matrix decomposition method.
4. The travel recommendation method based on the collaborative memory network as claimed in claim 1 or 2, wherein in step 5, neighborhood structure information o about user uuiComprises the following steps:
in the formula, puivWeight coefficient representing user u, cvRepresenting a user feature vector of each user v in a user external memory matrix in a user neighborhood set about the sight point i, n (i) representing a user neighborhood set formed by all users visiting the sight point i, u being 1,2, …, n, n representing the number of the users; i is 1,2, …, m, m indicates the number of sights.
5. The method as claimed in claim 1 or 2, wherein in step 7, the neighborhood structure information o of the scenic spot i is obtainediuComprises the following steps:
in the formula, piujWeight coefficient, y, representing the sight ijS (u) represents a scenery feature vector in a scenery external memory matrix of each scenery j in a scenery neighborhood set of the user u, wherein u is 1,2, …, n, and n represents the number of users; i is 1,2, …, m, m indicates the number of sights.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010618323.2A CN111882381B (en) | 2020-06-30 | 2020-06-30 | Travel recommendation method based on collaborative memory network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010618323.2A CN111882381B (en) | 2020-06-30 | 2020-06-30 | Travel recommendation method based on collaborative memory network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111882381A true CN111882381A (en) | 2020-11-03 |
| CN111882381B CN111882381B (en) | 2022-06-03 |
Family
ID=73156626
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010618323.2A Active CN111882381B (en) | 2020-06-30 | 2020-06-30 | Travel recommendation method based on collaborative memory network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111882381B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112712418A (en) * | 2021-03-25 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Method and device for determining recommended commodity information, storage medium and electronic equipment |
| CN112749345A (en) * | 2021-02-09 | 2021-05-04 | 上海海事大学 | K nearest neighbor matrix decomposition recommendation method based on neural network |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180103052A1 (en) * | 2016-10-11 | 2018-04-12 | Battelle Memorial Institute | System and methods for automated detection, reasoning and recommendations for resilient cyber systems |
| US20190073363A1 (en) * | 2017-09-07 | 2019-03-07 | Xerox Corporation | Contextual memory bandit for proactive dialogs |
| CN110008408A (en) * | 2019-04-12 | 2019-07-12 | 山东大学 | A kind of session recommended method, system, equipment and medium |
| CN110263256A (en) * | 2019-06-21 | 2019-09-20 | 西安电子科技大学 | Personalized recommendation method based on multi-modal heterogeneous information |
| CN110413844A (en) * | 2019-05-24 | 2019-11-05 | 浙江工业大学 | Dynamic link prediction technique based on space-time attention depth model |
| CN110825884A (en) * | 2019-11-05 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Embedded representation processing method and device based on artificial intelligence and electronic equipment |
| CN110826700A (en) * | 2019-11-13 | 2020-02-21 | 中国科学技术大学 | Method for realizing and classifying bilinear graph neural network model for modeling neighbor interaction |
| CN111061856A (en) * | 2019-06-06 | 2020-04-24 | 北京理工大学 | Knowledge perception-based news recommendation method |
| CN111222054A (en) * | 2020-01-03 | 2020-06-02 | 中国计量大学 | Session social contact recommendation method based on context neighbor relation modeling |
-
2020
- 2020-06-30 CN CN202010618323.2A patent/CN111882381B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180103052A1 (en) * | 2016-10-11 | 2018-04-12 | Battelle Memorial Institute | System and methods for automated detection, reasoning and recommendations for resilient cyber systems |
| US20190073363A1 (en) * | 2017-09-07 | 2019-03-07 | Xerox Corporation | Contextual memory bandit for proactive dialogs |
| CN110008408A (en) * | 2019-04-12 | 2019-07-12 | 山东大学 | A kind of session recommended method, system, equipment and medium |
| CN110413844A (en) * | 2019-05-24 | 2019-11-05 | 浙江工业大学 | Dynamic link prediction technique based on space-time attention depth model |
| CN111061856A (en) * | 2019-06-06 | 2020-04-24 | 北京理工大学 | Knowledge perception-based news recommendation method |
| CN110263256A (en) * | 2019-06-21 | 2019-09-20 | 西安电子科技大学 | Personalized recommendation method based on multi-modal heterogeneous information |
| CN110825884A (en) * | 2019-11-05 | 2020-02-21 | 腾讯科技（深圳）有限公司 | Embedded representation processing method and device based on artificial intelligence and electronic equipment |
| CN110826700A (en) * | 2019-11-13 | 2020-02-21 | 中国科学技术大学 | Method for realizing and classifying bilinear graph neural network model for modeling neighbor interaction |
| CN111222054A (en) * | 2020-01-03 | 2020-06-02 | 中国计量大学 | Session social contact recommendation method based on context neighbor relation modeling |
Non-Patent Citations (1)
| Title |
|---|
| 陈细玉等: "一种融合注意力和记忆网络的序列推荐算法", 《信息技术与网络安全》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112749345A (en) * | 2021-02-09 | 2021-05-04 | 上海海事大学 | K nearest neighbor matrix decomposition recommendation method based on neural network |
| CN112749345B (en) * | 2021-02-09 | 2023-11-14 | 上海海事大学 | K neighbor matrix decomposition recommendation method based on neural network |
| CN112712418A (en) * | 2021-03-25 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Method and device for determining recommended commodity information, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111882381B (en) | 2022-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110555112B (en) | Interest point recommendation method based on user positive and negative preference learning | |
| CN109977283B (en) | Tourism recommendation method and system based on knowledge graph and user footprint | |
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| CN110287336B (en) | Tourist map construction method for tourist attraction recommendation | |
| CN109785062B (en) | Hybrid neural network recommendation system based on collaborative filtering model | |
| CN110288436A (en) | A kind of personalized recommending scenery spot method based on the modeling of tourist's preference | |
| CN112380435A (en) | Literature recommendation method and recommendation system based on heterogeneous graph neural network | |
| CN105069717A (en) | Personalized travel route recommendation method based on tourist trust | |
| CN113505204B (en) | Recall model training method, search recall device and computer equipment | |
| CN111882381B (en) | Travel recommendation method based on collaborative memory network | |
| CN110795571A (en) | Cultural tourism resource recommendation method based on deep learning and knowledge graph | |
| CN111008336A (en) | Content recommendation method, device and equipment and readable storage medium | |
| CN113744032B (en) | Book recommendation method, related device, equipment and storage medium | |
| CN112800292A (en) | Cross-modal retrieval method based on modal specificity and shared feature learning | |
| CN113486190A (en) | Multi-mode knowledge representation method integrating entity image information and entity category information | |
| CN115408621B (en) | Interest point recommendation method considering auxiliary information characteristic linear and nonlinear interaction | |
| CN110532464B (en) | Tourism recommendation method based on multi-tourism context modeling | |
| Beni et al. | A fast module identification and filtering approach for influence maximization problem in social networks | |
| CN112800111B (en) | Position prediction method based on training data mining | |
| CN110008411A (en) | It is a kind of to be registered the deep learning point of interest recommended method of sparse matrix based on user | |
| CN116992151A (en) | Online course recommendation method based on double-tower graph convolution neural network | |
| CN115982652A (en) | Cross-modal emotion analysis method based on attention network | |
| CN115910232A (en) | Multi-view drug pair response prediction method, device, equipment and storage medium | |
| Yu et al. | Personalized recommendation of collective points-of-interest with preference and context awareness | |
| Liu | Research on personalized minority tourist route recommendation algorithm based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |