CN105069717A - Personalized travel route recommendation method based on tourist trust - Google Patents

Abstract

The present invention relates to a personalized travel route recommendation method based on tourist trust. Firstly, photo information with a geographical label and historical tourist information in a network are gathered, a large number of photo information is subjected to preprocessing, reliable interest point information is obtained, then a body database is constructed by using body-based modeling thought, interest point probability is predicted by using the mixed model of Markov and an subject, the travel route generation algorithm based on interest point heat is generated, and finally combined with a user trust weighted tourist route, a final route is recommended to a user. According to the method, the real travel information of users in a social network is fully utilized, a personalized travel route recommendation service can be effectively provided to the user, and the method has a good reference value for transportation service departments and travel agencies.

Description

A kind of personalized travelling route recommend method based on visitor's degree of belief

Technical field

The present invention relates to Internet technical field, specifically based on social networks and a kind of personalized travelling route recommend method based on visitor's degree of belief of data mining design.

Background technology

Social media website develops rapidly in recent years, and external Panoramio, Flickr and domestic bean cotyledon, Sina's microblogging etc. provide platform for people's information interchange.People like by word, and the every aspect, particularly tour schedule of the form records such as picture oneself life, people are very willing to share journey what is seen and heard.Traditional travel agency just experiences according to the tourism of masses or browses the modes such as tourism blog daily record and plans circuit, and usually more consuming time, do not make full use of the advantages such as infotech, can not meet the diversified demand of user, result is also unsatisfactory.Tour site then stresses focus recommendation, popular recommendation and various ticketing service purchase activity.

At present in social networks, most tourists usually can add certain contact person according to personal interest hobby or add certain interest group, and these can be household, friend or celebrity etc.Famous American fact-finding organ has been investigated and has been affected user and believe certain factor of recommending, and the user of result display 90% believes the recommendation of their friend.That is, the degree of belief between user affects the selection of user travelling route to a certain extent.

Tourist industry is as the mainstay industry of the national economic development, and personalized tourist service is recommended also to receive very large attention.The arrival of large data makes per secondly all can produce hundreds of data, and from mass data, how to excavate advantageous information is a problem demanding prompt solution.Personalized recommendation has become the major technique solving problem of information overload, is widely used in fields such as film, music and ecommerce, and major part adopts single recommended technology, does not take into full account the preference of user, also there is the problems such as Deta sparseness simultaneously.

Summary of the invention

For above weak point, the invention provides a kind of by obtaining data message of travelling really from social media website, in conjunction with point of interest temperature and users to trust degree, the personalized travelling route recommend method based on visitor's degree of belief of variation, personalized travelling route advisory opinion can be provided for user.

To achieve these goals, the technical solution adopted in the present invention is made up of following steps:

(1) call photo and share community website API, crawl with the tourism photographic intelligence collection of the geographical labels history tourist information corresponding with photo according to geographical longitude and latitude border, and utilize photographic intelligence to excavate tour interest point;

(2) screen the essential information of history visitor, the determined tour interest point of integrating step (1), adopt the method for Ontology Modeling to build ontology database, comprise user model and point of interest model;

(3) according to the travelling route requirement of user's input, the ontology database of coupling step (2), adopts the travelling route generating algorithm based on point of interest temperature to generate candidate line collection;

(4) degree of belief between the history visitor calculating corresponding point of interest in any one candidate line that user and candidate line of playing concentrate according to the user model of step (2), obtain user to the average degree of belief of this candidate line and weighting process, obtain recommending travelling route and being presented to user.

The method of photographic intelligence data mining tour interest point is utilized to be realized by following steps in above-mentioned steps (1):

A () adopts Shannon entropy method comparison film information data to carry out pre-service;

B () is according to geographical label information cluster and form tour interest point;

C () tour interest point is named.

The Shannon entropy method of above-mentioned steps (a) is specifically:

$E\left(u\right)=-\underset{m}{\overset{M\left(u\right)}{\Σ}}{P}_m\left(u\right){\mathrm{logP}}_m\left(u\right)$

$P_m\left(u\right)=\frac{D_m\left(u\right)}{\underset{m}{\overset{M\left(u\right)}{\Σ}}{D}_m\left(u\right)}$

Wherein, u is user, and E (u) is Shannon entropy, P _mu () is the probability that user u took pictures in the m month, D _mu () is the number of pictures of user u in the tour interest point m month, M (u) is that user u takes pictures at tour interest point the set of month m, and E (u) is larger, and user is that the probability of local is larger.

Above-mentioned user model comprises user ID, sex, age, the time of going on a tour, current location, preference sight spot type, good friend ID; Point of interest model comprises interest point name, point of interest type, travel time, weather condition, plays the residence time.

The concrete grammar of above-mentioned steps (3) is:

(3.1) according to the travelling route requirement of user's input, the ontology database of coupling step (2);

(3.2) probability of each tour interest point of playing is predicted according to the mixture model of current location and history point of interest record Markov model and topic model;

(3.3) the larger tour interest point of probability is chosen stored in queue, and using current location as starting point;

(3.4) calculate the time of the current line that point is formed from starting point to tour interest, if this wire time is less than free time, then perform step (3.5); Otherwise, export current line and be alternative circuit, perform step (3.6);

(3.5) according to each tour interest point probability of step (3.2) gained, the tour interest point that probability of being played by k before in queue is larger adds in current line respectively, then returns step (3.4);

(3.6) temperature of each tour interest point in alternative circuit is calculated;

(3.7) temperature of temperature mean value as this alternative circuit of each point of interest in alternative circuit is got, the alternative circuit alternatively sets of lines that before selecting, k temperature is larger.

The Markov model of above-mentioned steps (3.2) and the mixture model of topic model are specially:

$P\left(l_i\right|l_{i-1,}{h}^u)=\frac{P\left(l_i\right|l_{i-1})}{C(l_{i-1},h^u)}\frac{P\left(l_i\right|h^u)}{P\left(l_i\right)}$

Wherein, l _i, l _i-1for tour interest point, h ^ufor the history point of interest record of user u, P (l _i| l _i-1) be tour interest point l _i-1to l _itransition probability, P (l _i| h ^u) be predict according to the history point of interest record of user u the point of interest l that plays _iprobability, P (l _i) be the l that plays in all history point of interest records _iprobability, C (l _i-1, h ^u) represent normalization factor, Uni-Gram recall rate.

The computing method of point of interest temperature in above-mentioned steps (3.6):

$H\left(l_i\right)=0.5\×\frac{\underset{u_i\∈U}{\Σ}{\mathrm{VT}}_{u_i,l_i}}{\underset{u_i\∈U,l_i\∈L}{\Σ}{\mathrm{VT}}_{u_i,l_i}}+0.5\×{\mathrm{rank}}_{l_i}$

Wherein: H (l _i) represent tour interest point l _itemperature, temperature value is between 0 ~ 1, and it is more welcome to be worth this tour interest point of larger expression; represent user u _iplay this tour interest point l _inumber of times; L is the set of tour interest point, L={l _i; U is the set of all users, U={u _i; Rank _lifor the final ranking of tour interest point;

computing formula is as follows:

Above-mentioned users to trust degree computing method are:

User u _ato user u _bdegree of belief be adopt the degree of belief computing formula based on PageRank algorithm:

Wherein, u _jfor user u _agood friend, u _aall good friends be { u ₁, u ₂, u _b..., u _j..., u _n, PR (u _j) represent good friend u _jpageRank value.

Personalized travelling route recommend method based on visitor's degree of belief of the present invention utilizes group intelligence to obtain data message of travelling really from social media website, analyze its tourism dynamically, in conjunction with users to trust degree and point of interest temperature, for user provides variation, personalized travelling route advisory opinion, compared with existing technology, the invention has the advantages that:

1) from a large amount of from the true tourism data mining tour interest point social networks, be better than the popular artificial tour interest point arranged, have more objectivity.

2) realize personalized recommendation based on User-ontology model (essential information and history point of interest record) and users to trust degree, adding users to the degree of belief of recommendation results, instead of adopts popular score information.

3) consider the information such as current location, type of preferences and free time, meet users on diversity, personalized travelling route is provided.

4) adopt the probability predicting following tour interest point of playing based on Markov and theme mixture model, improve the objectivity and degree of accuracy of recommending travelling route, promote the satisfaction of client;

5) in conjunction with point of interest temperature, mixing point of interest forecast model and free time limit design rational travelling route generating algorithm, improve and recommend performance, for tourism planning and tour itineraries exploitation etc. provide reference frame, promote the development of tourist industry.

Accompanying drawing explanation

Fig. 1 is the personalized travelling route recommend method frame diagram based on visitor's degree of belief;

Fig. 2 is that tour interest point excavates process flow diagram;

Fig. 3 is user model frame diagram;

Fig. 4 is point of interest model framework figure.

Specific implementation method

Below in conjunction with accompanying drawing, technical scheme of the present invention is further illustrated.

As shown in Figure 1, the personalized travelling route recommend method based on visitor's degree of belief of the present embodiment is specifically realized by following steps:

(1) call Panoramio photo and share community website API, crawl history tourist information corresponding with photo with the tourism photographic intelligence collection of geographical labels in region, tourist attractions according to geographical longitude and latitude border, and utilize photographic intelligence to excavate tour interest point;

(1.1) crawl travel information collection, mainly comprise history visitor essential information and picture data collection.

Visitor's essential information, comprises visitor ID, sex, age and good friend ID;

Photographic intelligence collection, comprises photo ID, visitor ID, latitude, longitude, label and time;

(1.2) tour interest point excavates, specifically:

As shown in Figure 2, the excavation process flow diagram of tour interest point and popular Tour region, obtains some row points of interest L={l by technology such as pre-service, cluster and statistics ₁, l ₂..., l _i..., specifically:

A () adopts Shannon entropy method comparison film information data to carry out pre-service, filter resident's photo that non-tourism is relevant;

Exist in the data of collecting as wedding according to, birthday photograph, the scene of the accident etc. and the incoherent photo of tourism, these are recommended without any contributed value for tourism, therefore need the picture data of non-tourism to reject, and improve tourism and recommend performance and efficiency.

The present invention utilizes Shannon entropy method to distinguish life of urban resident photo, general tourists a city tour the residence time about one week, in the same moon or continuous at most 2 months, compare local resident, in 1 year, every month all likely takes pictures, therefore at this, a threshold epsilon is set, as E (u) > ε, we using this visitor as local resident, be abnormity point, should reject.

Shannon entropy specific algorithm is:

$E\left(u\right)=-\underset{m}{\overset{M\left(u\right)}{\Σ}}{P}_m\left(u\right){\mathrm{logP}}_m\left(u\right)$

$P_m\left(u\right)=\frac{D_m\left(u\right)}{\underset{m}{\overset{M\left(u\right)}{\Σ}}{D}_m\left(u\right)}$

Wherein, u is user, and E (u) is Shannon entropy, P _mu () is the probability that user u took pictures in the m month, D _mu () is the number of pictures of user u in the tour interest point m month, M (u) is the set of user u at the month m that takes pictures of tour interest point, and E (u) is larger, and user is that the probability of local is larger.

What information entropy was the most frequently used is with 2 the end of for, and unit is bit (bit); Other the end and unit can also be adopted, and can exchange.Take herein with 2 the end of as.

B () is according to geographical label information cluster and form tour interest point;

Utilize the longitude and latitude label information of average drifting (MeanShift) the clustering algorithm comparison film of density based to carry out cluster and form class bunch, thus determine tour interest point.

Average drifting be one non-parametric, specifically:

Known given picture location p ₀, its Meanshift vector m _w,G(p ₀) computing formula is:

$m_{w,G}\left(p_0\right)=\frac{{\Σ}_{i=1}^n{p}_{i}g\left(\right|\left|\right(p_0-p_i)/w|{|}^2)}{{\Σ}_{i=1}^{n}g\left(\right|\left|\right(p_0-p_i)/w|{|}^2)}-p_0$

Wherein, g represents the weight of particular core function G corresponding to each location point, general, and g is the negative direction of kernel function G differentiate, and w is a frequency range parameter (radius), p _ibe i-th picture location.

Mean shift procedures is the place set of a series of certain position of sensing, meets:

p _i' ₊₁＝p _i+m _w,G(p _i)

This is an iterative process, until Meanshift vector m _w,G(p _i) convergence, stop iteration, so just obtain an interest point information, find a series of interest point set L={l by that analogy _i, l _i={ p _i.

C () tour interest point is named

Adopt n-gram model (n-grams) statistical method to calculate the frequency of all text labels in each class bunch, finally select text label that frequency is the highest as interest point name.

N-gram model is the algorithm adding up word frequency in the text, and in n-gram model, a continuous print character string sequence can be regarded as in a sentence, namely can be individual character sequence, also can be word sequence.One of effect of n-gram is exactly the probability that prediction word sequence occurs.Utilize large-scale corpus and ripe n-gram model, higher cutting accuracy can be obtained easily.Oneself is through there are some researches show, use three metagrammars (n=3), when not considering unknown word, cutting accuracy is up to more than 98%.

(2) screen the essential information of history visitor, the determined tour interest point of integrating step (1), according to Ontology Modeling thought, use for reference " seven footworks " thought, build ontology database, comprise user model and point of interest model.

As shown in Figure 3, user model comprises user ID, sex, age, the time of going on a tour, current location, preference sight spot type, good friend ID.

As shown in Figure 4, point of interest model comprises interest point name, point of interest type, travel time, weather condition, plays the residence time.

Association between final two models can be expressed as user and to play certain point of interest.

Wherein point of interest type mainly contains ruins, historic site, museum, theme park, landscape and temple etc.;

Travel time is the mean value according to photo temporal information in point of interest class bunch;

Residence time of playing is difference according to photo temporal information maxima and minima in point of interest class bunch.

(3) according to the travelling route requirement of user's input, the ontology database of coupling step (2), adopts the travelling route generating algorithm based on point of interest temperature to generate candidate line.

(3.1) according to the travelling route requirement of user's input, as: current location, point of interest type preference, free time etc., conventionally mate the ontology database of step (2);

(3.2) predict the probability of other tour interest points of playing according to current location and history point of interest record, mainly adopt the mixture model of Markov model and topic model to predict tour interest point probability, specific formula for calculation is:

$P\left(l_i\right|l_{i-1,}{h}^u)=\frac{P\left(l_i\right|l_{i-1})}{C(l_{i-1},h^u)}\frac{P\left(l_i\right|h^u)}{P\left(l_i\right)}$

Wherein, l _i, l _i-1for tour interest point, h ^ufor the history point of interest record of user u, P (l _i| l _i-1) be tour interest point l _i-1to l _iprobability, P (l _i| h ^u) be predict according to the history point of interest record of user u the point of interest l that plays _iprobability, P (l _i) be the l that plays in all history point of interest records _iprobability, C (l _i-1, h ^u) represent normalization factor, Uni-Gram recall rate.

(3.3) the tour interest point that select probability is larger stored in queue, and using current location as starting point;

(3.4) calculate the time of the current line that point is formed from starting point to tour interest, if this wire time is less than free time, then perform step (3.5); Otherwise, export current line and be alternative circuit, perform step (3.6);

(3.5) transition probability of the different tour interest points obtained according to step (3.2), adds in current line respectively by tour interest point larger for k probability of playing before in queue, then returns and perform step (3.4);

(3.6) calculate the temperature of each point of interest in alternative circuit, computing formula is as follows:

$H\left(l_i\right)=0.5\×\frac{\underset{u_i\∈U}{\Σ}{\mathrm{VT}}_{u_i,l_i}}{\underset{u_i\∈U,l_i\∈L}{\Σ}{\mathrm{VT}}_{u_i,l_i}}+0.5\×{\mathrm{rank}}_{l_i}$

Wherein: H (l _i) represent tour interest point l _itemperature, temperature value is between 0 ~ 1, and it is more welcome to be worth this tour interest point of larger expression; represent user u _iplay this tour interest point l _inumber of times; L is the set of tour interest point, L={l _i; U is the set of all users, U={u _i; for the final ranking of tour interest point;

computing formula is as follows:

(3.7) temperature of temperature mean value as this alternative circuit of each point of interest in alternative circuit is got, the alternative circuit alternatively sets of lines that before selecting, k temperature is larger.

By a series of tour interest point composition line sequence, add that point of interest average temperature rank obtains the larger alternative circuit of a front k temperature alternatively travelling route collection S={s ₁, s ₂..., s _i..., s _k, represent circuit s _ia middle jth point of interest; N represents circuit s _itotal number of middle point of interest, i, j, N are positive integer.The hot value of each circuit is:

(4) degree of belief between user and the history visitor of corresponding point of interest in this candidate line of playing is calculated according to the user model of step (2), obtain user to the average degree of belief of this candidate line and weighting process, obtain recommending travelling route and being presented to user.

Degree of belief between user, by building user-user social contact matrix, utilizes PageRank algorithmic technique to analyze the PageRank value of each user, is technorati authority, then utilizes technorati authority to calculate the degree of belief of user to point of interest trip player each in candidate line.

Known users u _awith user u _b, be then shown below user u _ato user u _bdegree of belief be final degree of belief affects the sequence of recommendation list as weights.

Wherein, u _jfor user u _agood friend, u _aall good friends be { u ₁, u ₂, u _b..., u _j..., u _n, PR (u _j) represent good friend u _jpageRank value.

The degree of belief of user to all history visitors of point of interest each in candidate line is obtained according to degree of belief computing formula, again the degree of belief of all history visitors of each point of interest is added and, be the degree of belief of user to point of interest each in travelling route, the degree of belief of all points of interest added and gets its mean value as the degree of belief of user to this candidate travelling route then the up-to-date score value obtaining every bar circuit is score (s _i) ', score (s the most at last _i) ' present to user according to the result of descending sort as travelling route recommendation list.

In a word, the present invention according to the current location of user, free time, and in conjunction with user characteristics and history point of interest record, can recommend travelling route neatly.

Be more than the preferred embodiment describing a kind of personalized travelling route recommend method based on visitor's degree of belief by reference to the accompanying drawings in detail, be not used for limiting the scope of the present invention.

Claims (8)

1., based on a personalized travelling route recommend method for visitor's degree of belief, it is characterized in that being made up of following steps:

(1) call photo and share community website API, crawl with the tourism photographic intelligence collection of the geographical labels history tourist information corresponding with photo according to geographical longitude and latitude border, and utilize photographic intelligence to excavate tour interest point;

(2) screen the essential information of history visitor, the determined tour interest point of integrating step (1), adopt the method for Ontology Modeling to build ontology database, comprise user model and point of interest model;

(3) according to the travelling route requirement of user's input, the ontology database of coupling step (2), adopts the travelling route generating algorithm based on point of interest temperature to generate candidate line collection;

(4) degree of belief between the history visitor calculating corresponding point of interest in any one candidate line that user and candidate line of playing concentrate according to the user model of step (2), obtain user to the average degree of belief of this candidate line and weighting process, obtain recommending travelling route and being presented to user.

2. the personalized travelling route recommend method based on visitor's degree of belief according to claim 1, is characterized in that: utilize the method for photographic intelligence data mining tour interest point to be realized by following steps in step (1):

A () adopts Shannon entropy method comparison film information data to carry out pre-service;

B () is according to geographical label information cluster and form tour interest point;

C () tour interest point is named.

3. the personalized travelling route recommend method based on visitor's degree of belief according to claim 2, is characterized in that: the Shannon entropy method of described step (a) specifically:

$E\left(u\right)=-\underset{m}{\overset{M\left(u\right)}{\Σ}}{P}_m\left(u\right){\mathrm{logP}}_m\left(u\right)$

$P_m\left(u\right)=\frac{D_m\left(u\right)}{\underset{m}{\overset{M\left(u\right)}{\Σ}}{D}_m\left(u\right)}$

Wherein, u is user, and E (u) is Shannon entropy, P _mu () is the probability that user u took pictures in the m month, D _mu () is the number of pictures of user u in the tour interest point m month, M (u) is that user u takes pictures at tour interest point the set of month m, and E (u) is larger, and user is that the probability of local is larger.

4. the personalized travelling route recommend method based on visitor's degree of belief according to claim 1 or 2 or 3, is characterized in that: described user model comprises user ID, sex, age, the time of going on a tour, current location, preference sight spot type, good friend ID; Point of interest model comprises interest point name, point of interest type, travel time, weather condition, plays the residence time.

5. the personalized travelling route recommend method based on visitor's degree of belief according to claim 4, is characterized in that: the concrete grammar of described step (3) is:

(3.1) according to the travelling route requirement of user's input, the ontology database of coupling step (2);

(3.2) probability of each tour interest point of playing is predicted according to the mixture model of current location and history point of interest record Markov model and topic model;

(3.3) the larger tour interest point of probability is chosen stored in queue, and using current location as starting point;

(3.4) calculate the time of the current line that point is formed from starting point to tour interest, if this wire time is less than free time, then perform step (3.5); Otherwise, export current line and be alternative circuit, perform step (3.6);

(3.5) according to each tour interest point probability of step (3.2) gained, the tour interest point that probability of being played by k before in queue is larger adds in current line respectively, then returns step (3.4);

(3.6) temperature of each tour interest point in alternative circuit is calculated;

(3.7) temperature of temperature mean value as this alternative circuit of each point of interest in alternative circuit is got, the alternative circuit alternatively sets of lines that before selecting, k temperature is larger.

6. the personalized travelling route recommend method based on visitor's degree of belief according to claim 5, is characterized in that: the Markov model of described step (3.2) and the mixture model of topic model are specially:

$P\left(l_i\right|l_{i-1,}{h}^u)=\frac{P\left(l_i\right|l_{i-1})}{C(l_{i-1},h^u)}\frac{P\left(l_i\right|h^u)}{P\left(l_i\right)}$

Wherein, l _i, l _i-1be tour interest point, h ^ufor the history point of interest record of user u, P (l _i| l _i-1) be tour interest point l _i-1to l _itransition probability, P (l _i| h ^u) be predict according to the history point of interest record of user u the point of interest l that plays _iprobability, P (l _i) be the l that plays in all history point of interest records _iprobability, C (l _i-1, h ^u) represent normalization factor, Uni-Gram recall rate.

7. the personalized travelling route recommend method based on visitor's degree of belief according to claim 5, is characterized in that: the computing method of point of interest temperature in described step (3.6):

$H\left(l_i\right)=0.5\×\frac{\underset{u_i\∈U}{\Σ}{\mathrm{VT}}_{u_i,l_i}}{\underset{u_i\∈U,l_i\∈L}{\Σ}{\mathrm{VT}}_{u_i,l_i}}+0.5\×{\mathrm{rank}}_{l_i}$

Wherein: H (l _i) represent tour interest point l _itemperature, temperature value is between 0 ~ 1, and it is more welcome to be worth this tour interest point of larger expression; represent user u _iplay this tour interest point l _inumber of times; L is the set of tour interest point, L={l _i; U is the set of all users, U={u _i; for the final ranking of tour interest point;

computing formula is as follows:

8. the personalized travelling route recommend method based on visitor's degree of belief according to claim 1, is characterized in that: described users to trust degree computing method are:

User u _ato user u _bdegree of belief be adopt the degree of belief computing formula based on PageRank algorithm:

Wherein, u _jfor user u _agood friend, u _aall good friends be { u ₁, u ₂, u _b..., u _j..., u _n, PR (u _j) represent good friend u _jpageRank value.