Hybrid rearrangement tourism recommendation method and system based on multi-objective optimization

Abstract

The present invention discloses a hybrid rearrangement tourism recommendation method based on multi-objective optimization, comprising the following steps: S1: Dataset construction: Collect historical visit sequences of users, and according to the time order of users' visits, take the last visited place as a test set, the second to last visited place as a validation set, and the rest as training sets; S2: Construction of a multi-objective optimization hybrid rearrangement model: including population initialization, calculation of objective function and population evolution; S3: Model training: Use the training set to train the multi-objective optimization hybrid rearrangement model, and use the validation set to adjust the model parameters to obtain the optimal model; S4: Model prediction: Use the optimal model to predict the users in the test set and obtain the travel recommendation list recommended to the users. A hybrid rearrangement tourism recommendation system based on multi-objective optimization is also disclosed. The present invention can improve the diversity of the entire system while ensuring the accuracy of the entire recommendation system.

Description

Technical Field

[0001] The invention relates to the field of optimization calculation, and in particular to a hybrid rearrangement tourism recommendation method based on multi-objective optimization and a system thereof. Background Art

[0002] With the improvement of residents' living standards, residents' demands for spiritual culture and a better life are becoming stronger and stronger. Tourism has become an important part of people's leisure and entertainment, and the tourism industry has developed rapidly. Faced with huge and complex tourism data, people are easily drowned in a large amount of data, and it is difficult for users to quickly find the tourist attractions they need. Therefore, in order to meet people's increasingly urgent needs for efficient and intelligent tourism information, effectively solve tourism recommendation services, and recommend more favorite attractions to users, an efficient tourism recommendation system is very necessary. Usually, the effect of a single recommendation algorithm in solving practical problems is often not ideal. Hybrid recommendation algorithms can solve this problem. Each recommendation technology in the hybrid recommendation algorithm can make up for the shortcomings of each other. However, traditional recommendation algorithms only focus on improving the accuracy of the entire recommendation system, but often ignore the diversity of the entire system. Therefore, we should not only actively explore the interests of users but also pay attention to the diversity of recommendations. In order to solve the conflict between these two goals, a multi-objective optimization algorithm can be used to deal with the optimization problem between multiple conflicting goals.

[0003] Therefore, there is an urgent need to provide a new hybrid rearrangement tourism recommendation method and system based on multi-objective optimization to solve the above problems. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a hybrid rearrangement tourism recommendation method and system based on multi-objective optimization, which can improve the diversity of the entire system while ensuring the accuracy of the entire recommendation system.

[0005] In order to solve the above technical problems, a technical solution adopted by the present invention is to provide a hybrid rearrangement tourism recommendation method based on multi-objective optimization, comprising the following steps:

[0006] S1: Dataset construction: collect the user's historical access sequence and sort it according to the user's access ID and access time, filter the collected data set, and finally divide the data set. According to the user's time access sequence, the last access location is used as the test set, the second to last access location is used as the validation set, and the rest are used as the training set;

[0007] S2: Construction of multi-objective optimization hybrid rearrangement model: including population initialization, calculation of objective function and population evolution. The objective function is calculated by multi-objective optimization combined with MMR algorithm.

[0008] S3: Model training: Use the training set to train the multi-objective optimization hybrid rearrangement model constructed in step S2, and use the validation set to adjust the model parameters to obtain the optimal model;

[0009] S4: Model prediction: Use the optimal model obtained in step S3 to predict the users in the test set and obtain the travel recommendation list recommended to the users.

[0010] In a preferred embodiment of the present invention, in step S1, the collected data set is subjected to data screening, including filtering out locations visited by less than or equal to 10 different users and users who visit less than or equal to 10 different locations.

[0011] In a preferred embodiment of the present invention, the specific steps of step S2 include:

[0012] S201: Population initialization: Encode the weights α, β, γ of the three baseline recommendation algorithms and the weight λ in the rearrangement algorithm as an individual in the population, such as [α, β, γ, λ], randomly initialize the population with N individuals, generate reference points uniformly distributed on the unit target dimension, and generate a neighbor set;

[0013] S202: Calculation of objective function: Select user-based collaborative filtering (User_CF), item-based collaborative filtering (Item_CF), and SASRec as three mixed baseline recommendation algorithms, input the filtered data sets into the three baseline recommendation algorithms to obtain the three users' rating ranking lists for unknown items, which are and R SASRec , the three rating ranking lists are mixed linearly weighted to obtain the rating ranking list, with weights α, β, and γ respectively, and the user's score for each item is obtained as follows: Then, the score ranking list is input into the MMR algorithm for rearrangement to obtain the rearranged recommendation list. The weight in the MMR algorithm is λ. Finally, the objective function value is calculated based on the rearranged user recommendation list.

[0014] S203: Population evolution: Use the population evolution strategy in the MOEA / D algorithm to optimize individuals, where the crossover operator is a binary crossover operator and the mutation operator is a polynomial mutation operator. Through t generations of population evolution, the final recommendation list is obtained.

[0015] Furthermore, in step S202, the step of calculating the objective function includes:

[0016] (1) Accuracy function:

[0017] In the formula, R uis the recommendation list for user u, click is the last visited location in the user's historical visit sequence, {predictTOPN}∩{click} means that the last visited location in the user's historical visit sequence appears in the user's recommendation list, which is a hit, and |User| is the number of users in the entire data set;

[0018] (2) Diversity function:

[0019] In the formula, |R u | is the length of the recommendation list for user u, s(m,n) represents the similarity between two items m and n;

[0020] The final objective function is constructed as:

[0021]

[0022] Further, in step S202, the specific description of the MMR algorithm is:

[0023] The MMR algorithm combines the relevance and diversity of the ranking results into the following formula:

[0024]

[0025] In the formula, Q represents the user, R represents the recommendation score ranking list input into the algorithm, D represents the recommendation result set, S represents the selected set in R, R\S represents the unselected set in R, λ is the weight coefficient for adjusting the accuracy and diversity of the recommendation results, Sim2 is the similarity between items, Sim1 is a relevance score of the user to the item, calculated as follows:

[0026]

[0027] In order to solve the above technical problems, another technical solution adopted by the present invention is to provide a hybrid rearrangement tourism recommendation system based on multi-objective optimization, comprising:

[0028] The dataset construction module is used to collect the user's historical visit sequence from Foursquare and / or travel websites, sort them according to the user's visit ID and visit time, filter the collected dataset, and finally divide the dataset according to the user's time visit sequence, taking the last visited location as the test set, the second to last visited location as the validation set, and the rest as the training set;

[0029] A multi-objective optimization hybrid rearrangement model building module is used to build a multi-objective optimization hybrid rearrangement model;

[0030] The model training module is used to train the constructed multi-objective optimization hybrid rearrangement model using the training set, and adjust the model parameters using the validation set to obtain the optimal model;

[0031] The model prediction module is used to use the obtained optimal model to predict the users in the test set and obtain the travel recommendation list recommended to the users.

[0032] In a preferred embodiment of the present invention, the multi-objective optimization hybrid rearrangement model construction module includes a population initialization unit, an objective function calculation unit, and a population evolution unit. The objective function calculation unit uses multi-objective optimization combined with the MMR algorithm for calculation.

[0033] Furthermore, the population initialization unit is used for initializing the population, including initializing individuals, reference points and neighbor sets of the population.

[0034] Furthermore, the objective function calculation unit is used to input the data set into three baseline recommendation algorithms to obtain three users' rating ranking lists for unknown items, linearly weighted mix the three rating ranking lists, with the weights being the decision variables of the population, and then input the weighted rating ranking lists into the MMR algorithm for rearrangement to obtain a rearranged recommendation list and calculate the objective function value.

[0035] Furthermore, the population evolution unit is used to guide the evolution of the population using the population evolution strategy in MOEA / D, and finally obtain the user's recommendation list.

[0036] The beneficial effects of the present invention are:

[0037] (1) The present invention uses a multi-objective optimization method to optimize the accuracy and diversity of the tourism recommendation system and adds a re-ranking algorithm, which improves the diversity of recommendations while ensuring the accuracy of recommendations, greatly improving the user experience;

[0038] (2) The present invention adopts a multi-objective optimization hybrid rearrangement algorithm to make TOP-N recommendations for the attractions that the user is going to visit next, contributing to the development of subsequent recommendation algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a general framework flow chart of the hybrid rearrangement tourism recommendation method based on multi-objective optimization of the present invention;

[0040] Figure 2 is a schematic diagram of a recommended model of hybrid rearrangement for multi-objective optimization;

[0041] Figure 3 is a schematic diagram of the composition of the data set;

[0042] Figure 4 is the Pareto frontier surface comparison diagram of the HSC data set;

[0043] Figure 5 It is a Pareto front comparison diagram of the NYC data set;

[0044] Figure 6 It is a structural block diagram of the hybrid rearrangement tourism recommendation system based on multi-objective optimization. DETAILED DESCRIPTION

[0045] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0046] See also Figure 1 and Figure 2 , the embodiment of the present invention includes:

[0047] A hybrid rearrangement tourism recommendation method based on multi-objective optimization includes the following steps:

[0048] S1: Dataset construction: Collect users’ historical visit sequences from Foursquare and / or travel websites, sort them according to the user’s visit ID and visit time, filter the collected data sets, and finally divide the data sets. According to the user’s time visit sequence, the last visited location is used as the test set, the second to last visited location is used as the validation set, and the rest are used as the training set;

[0049] Furthermore, the collected data set is subjected to data screening, including filtering out a number of locations that are rarely visited by users and users who visit a few locations.

[0050] S2: Construction of multi-objective optimization hybrid rearrangement model: including population initialization, calculation of objective function and population evolution; specific steps include:

[0051] S201: Population initialization: MOEA / D is a general algorithm framework that decomposes the multi-objective optimization problem into multiple different single-objective sub-problems and uses an evolutionary algorithm to simultaneously optimize these sub-problems. In this algorithm, the decomposition method of the Chebyshev weight aggregation method is used, and its formula is:

[0052]

[0053] The weights α, β, γ of the three baseline recommendation algorithms and the weight λ in the rearrangement algorithm are encoded as an individual in the population such as [α, β, γ, λ], and the population is randomly initialized with the number of individuals N = 35. is the reference point set, where is the minimum value of each target in the current population, for each weight vector μ i (1≤i≤35), select from the current population i The individual with the closest Euclidean distance is taken as the neighbor individual, and finally the neighbor set N = {N1, N2, ..., N 35};

[0054] S202: Calculation of objective function: Select user-based collaborative filtering (User_CF), item-based collaborative filtering (Item_CF), and SASRec as three mixed baseline recommendation algorithms, input the filtered data sets into the three baseline recommendation algorithms to obtain the three users' rating ranking lists for unknown items, which are and R SASRec , the three rating ranking lists are mixed linearly weighted to obtain the rating ranking list, with weights α, β, and γ respectively, and the user's score for each item is obtained as follows: Then, the score ranking list is input into the MMR algorithm for rearrangement to obtain the rearranged recommendation list. The weight in the MMR algorithm is λ. Finally, the objective function value is calculated based on the rearranged user recommendation list.

[0055] (1) Accuracy function:

[0056] In the formula, R u is the recommendation list for user u, click is the last visited location in the user's historical visit sequence, {predictTOPN}∩{click} means that the last visited location in the user's historical visit sequence appears in the user's recommendation list, which is a hit, and |User| is the number of users in the entire data set;

[0057] (2) Diversity function:

[0058] In the formula, |R u | is the length of the recommendation list for user u, s(m,n) represents the similarity between two items m and n;

[0059] The final objective function is constructed as:

[0060]

[0061] The specific description of the MMR algorithm is as follows:

[0062] The MMR algorithm combines the relevance and diversity of the ranking results into the following formula:

[0063]

[0064] In the formula, Q represents the user, R represents the recommendation score ranking list input into the algorithm, D represents the recommendation result set, S represents the selected set in R, R\S represents the unselected set in R, λ is the weight coefficient for adjusting the accuracy and diversity of the recommendation results, Sim2 is the similarity between items, Sim1 is a relevance score of the user to the item, calculated as follows:

[0065]

[0066] S203: Population evolution: Use the population evolution strategy in the MOEA / D algorithm to optimize individuals, where the crossover operator is a binary crossover operator and the mutation operator is a polynomial mutation operator. Through t generations of population evolution, the final recommendation list is obtained.

[0067] S3: Model training: Use the training set to train the multi-objective optimization hybrid rearrangement model constructed in step S2, and use the validation set to adjust the model parameters to obtain the optimal model;

[0068] S4: Model prediction: Use the optimal model obtained in step S3 to predict the users in the test set and obtain the travel recommendation list recommended to the users.

[0069] This example experiment obtains the tourism information of Huangshan City on the Ctrip website through crawling technology. The content of the information includes user ID, attraction ID, time, rating and attraction coordinates. This information is sorted according to the user ID and the user's visit time, and the visit sequence of each user is obtained. Finally, the visit sequence of each user is combined into a tourism dataset called HSC. Secondly, in order to verify the universality of the model, the public dataset NYC in Foursquare is used for experiments. The NYC dataset includes user ID, location ID, user visit time, location, etc. In order to avoid cold start and data sparsity problems, the dataset is processed.

[0070] The data set for this experiment consists of two data sets: NYC and HSC. Data processing was performed on the HSC data set by filtering out locations visited by less than 4 different users and users who visited less than 4 different locations. Data processing was performed on the NYC data set by filtering out locations visited by less than 10 different users and users who visited less than 10 different locations. Finally, according to the user's time visit sequence, the last visited location was used as the test set, the second to last visited location was used as the validation set, and the rest were used as the training set. The composition of the data set used in this experiment is as follows: Figure 3 shown.

[0071] The two optimization goals are calculated based on the recommendation list recommended by the system for each user. The accuracy and diversity of the entire system represent the average of the accuracy and diversity of the recommendations for each user. Therefore, the accuracy and diversity of the entire system are calculated as follows:

[0072]

[0073]

[0074] Among them, |User| represents the number of users in the entire dataset. u is the recommendation list for user u, click is the last visited location in the user's historical visit sequence. {predictTOPN}∩{click} means that the last visited location in the user's historical visit sequence appears in the user's recommendation list, and it is a hit. Finally, the accuracy prediction probability of all users in the entire recommendation list is obtained. |R u | is the length of the recommendation list for user u, and s(m, n) represents the similarity between two items (i.e., attractions) m and n. Finally, the diversity prediction probability of all users in the entire recommendation list is obtained.

[0075] Table 1 shows the settings of some parameters of the multi-objective algorithm: the number of evolution generations is set to 20, the population size is set to 35, the number of objective functions is set to 2, the dimension of decision variables is set to 4, the crossover probability is set to 1, and the mutation probability is set to 1 / D.

[0076] Table 1

[0077]

[0078] In order to display the results as a whole, Figure 4 and Figure 5 The Pareto front surfaces obtained by hybrid recommendation based on other multi-objective algorithms on the HSC dataset and the NYC dataset are shown respectively. Each solution is an evaluation of the overall recommendation system and is also a balance between the two objective functions proposed by the present invention. From the overall trend of the entire Pareto front surface, it can be seen that the points on the front surface of the model of the present invention are almost all distributed above other algorithms, that is to say, it is superior to other models in terms of the diversity of overall recommendations. At the same time, the algorithm of the present invention also shows superior results in the hybrid rearrangement algorithm based on the MOEA / D algorithm, proving that adding the rearrangement algorithm can improve the effect of the recommendation system.

[0079] Tables 2 and 3 specifically show the performance results of hybrid recommendations based on other multi-objective algorithms. First of all, it can be seen from Tables 2 and 3 that compared with a variety of baseline recommendation algorithms (Item_CF, User_CF, SASRec, AttRec, BPR), this algorithm has shown good results in both evaluation indicators. In comparison with the hybrid (Vote) recommendation based on voting, the overall performance has also been greatly improved. In comparison with hybrid recommendations based on other multi-objective algorithms (MOEA / D, NSGA-Ⅲ, RVEA), good results are shown in both the best and average aspects, and good results are shown in the best, average and worst aspects of diversity. Therefore, compared with other models based on multi-objective optimization, the model proposed in the present invention has certain improvements in accuracy and diversity, greatly improving the user experience.

[0080] Table 2 HSC dataset

[0081]

[0082]

[0083] Table 3 NYC dataset

[0084]

[0085]

[0086] See also Figure 6 The example of the present invention also provides a hybrid rearrangement tourism recommendation system based on multi-objective optimization, including:

[0087] The dataset construction module is used to collect the user's historical visit sequence from Foursquare and / or travel websites, sort them according to the user's visit ID and visit time, filter the collected dataset, and finally divide the dataset according to the user's time visit sequence, taking the last visited location as the test set, the second to last visited location as the validation set, and the rest as the training set;

[0088] A multi-objective optimization hybrid rearrangement model building module is used to build a multi-objective optimization hybrid rearrangement model;

[0089] The model training module is used to train the constructed multi-objective optimization hybrid rearrangement model using the training set, and adjust the model parameters using the validation set to obtain the optimal model;

[0090] The model prediction module is used to use the obtained optimal model to predict the users in the test set and obtain the travel recommendation list recommended to the users.

[0091] Furthermore, the multi-objective optimization hybrid rearrangement model construction module includes a population initialization unit, an objective function calculation unit, and a population evolution unit.

[0092] The population initialization unit is used for initializing the population, including initializing individuals, reference points and neighbor sets of the population;

[0093] The objective function calculation unit is used to input the data set into three baseline recommendation algorithms to obtain three user rating ranking lists for unknown items, perform linear weighted mixing on the three rating ranking lists, with the weights being the decision variables of the population, and then input the weighted rating ranking lists into the MMR algorithm for rearrangement to obtain a rearranged recommendation list and calculate the objective function value;

[0094] The population evolution unit is used to guide the evolution of the population using the population evolution strategy in MOEA / D, and finally obtain the user's recommendation list.

[0095] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A hybrid rearrangement tourism recommendation method based on multi-objective optimization, characterized in that: The following steps are involved: S1: Dataset construction: collect the user's historical access sequence and sort it according to the user's access ID and access time, filter the collected data set, and finally divide the data set. According to the user's time access sequence, the last access location is used as the test set, the second to last access location is used as the validation set, and the rest are used as the training set; S2: Construction of multi-objective optimization hybrid rearrangement model: including population initialization, calculation of objective function and population evolution. The objective function is calculated by using user-based collaborative filtering (User_CF), item-based collaborative filtering (Item_CF), and SASRec as three baseline recommendation algorithms combined with MMR algorithm; S3: Model training: Use the training set to train the multi-objective optimization hybrid rearrangement model constructed in step S2, and use the validation set to adjust the model parameters to obtain the optimal model; S4: Model prediction: Use the optimal model obtained in step S3 to predict the users in the test set and obtain the travel recommendation list recommended to the users.

2. The hybrid rearrangement travel recommendation method based on multi-objective optimization according to claim 1 is characterized in that: In step S1 , the collected data set is subjected to data screening, including filtering out locations visited by less than or equal to 10 different users and users who visit less than or equal to 10 different locations.

3. The hybrid rearrangement travel recommendation method based on multi-objective optimization according to claim 1 is characterized in that: The specific steps of step S2 include: S201: Population initialization: Encode the weights α, β, γ of the three baseline recommendation algorithms and the weight λ in the rearrangement algorithm as an individual [α, β, γ, λ] in the population, randomly initialize the population with N individuals, generate reference points uniformly distributed on the unit target dimension and generate a neighbor set; S202: Calculation of objective function: Select user-based collaborative filtering (User_CF), item-based collaborative filtering (Item_CF), and SASRec as three mixed baseline recommendation algorithms, input the filtered data sets into the three baseline recommendation algorithms to obtain the three users' rating ranking lists for unknown items, which are , as well as , the three rating ranking lists are mixed linearly weighted to obtain the rating ranking list, with weights α, β, and γ respectively, and the user's score for each item is obtained as follows: ; Then input the score ranking list into the MMR algorithm to re-arrange it to obtain the re-arranged recommendation list, and the weight in the MMR algorithm is λ; finally, calculate the objective function value based on the re-arranged user recommendation list; S203: Population evolution: Use the population evolution strategy in the MOEA / D algorithm to optimize individuals, where the crossover operator is a binary crossover operator and the mutation operator is a polynomial mutation operator. Through t generations of population evolution, the final recommendation list is obtained.

4. The hybrid rearrangement travel recommendation method based on multi-objective optimization according to claim 3 is characterized in that: In step S202, the step of calculating the objective function includes: (1) Accuracy function: ; In the formula, is the recommendation list for user u, The last visited location in the user's historical visit sequence. It means that the last visited location in the user's historical visit sequence appears in the user's recommendation list, which is a hit. Represents the number of users in the entire dataset; (2) Diversity function: ； In the formula, The length of the recommended list for user u, Represents the similarity between two items m and n; The final objective function is constructed as: 。 5. The hybrid rearrangement travel recommendation method based on multi-objective optimization according to claim 3 is characterized in that: In step S202, the specific description of the MMR algorithm is: The MMR algorithm combines the relevance and diversity of the ranking results into the following formula: ； In the formula, Q represents the user, R represents the recommendation score ranking list input into the algorithm, D represents the recommendation result set, and S represents the set that has been selected in R. represents the unselected set in R, To adjust the weight coefficients of accuracy and diversity of recommendation results, is the similarity between items, is a relevance score for the user to the item, calculated as follows: 。 6. A hybrid rearrangement tourism recommendation system based on multi-objective optimization, characterized in that: include: The dataset construction module is used to collect the user's historical visit sequence from Foursquare and / or travel websites, sort them according to the user's visit ID and visit time, filter the collected dataset, and finally divide the dataset according to the user's time visit sequence, taking the last visited location as the test set, the second to last visited location as the validation set, and the rest as the training set; A multi-objective optimization hybrid rearrangement model construction module is used to construct a multi-objective optimization hybrid rearrangement model. The multi-objective optimization hybrid rearrangement model construction module includes a population initialization unit, an objective function calculation unit, and a population evolution unit. The objective function calculation unit uses user-based collaborative filtering (User_CF), item-based collaborative filtering (Item_CF), and SASRec as three mixed baseline recommendation algorithms combined with the MMR algorithm for calculation; The model training module is used to train the constructed multi-objective optimization hybrid rearrangement model using the training set, and adjust the model parameters using the validation set to obtain the optimal model; The model prediction module is used to use the obtained optimal model to predict the users in the test set and obtain the travel recommendation list recommended to the users.

7. The hybrid rearrangement tourism recommendation system based on multi-objective optimization according to claim 6 is characterized in that: The population initialization unit is used for initializing the population, including initializing individuals, reference points and neighbor sets of the population.

8. The hybrid rearrangement tourism recommendation system based on multi-objective optimization according to claim 6 is characterized in that: The objective function calculation unit is used to input the data set into three baseline recommendation algorithms to obtain three users' score ranking lists for unknown items, linearly weighted mix the three score ranking lists, with the weights being the decision variables of the population, and then input the weighted score ranking lists into the MMR algorithm for rearrangement to obtain the rearranged recommendation list and calculate the objective function value.

9. The hybrid rearrangement tourism recommendation system based on multi-objective optimization according to claim 6 is characterized in that: The population evolution unit is used to guide the evolution of the population using the population evolution strategy in MOEA / D, and finally obtain the user's recommendation list.

Images