CN111988744A - Position prediction method based on user moving mode - Google Patents

Abstract

The invention relates to a position prediction method based on a user's movement pattern, and belongs to the field of machine learning. The methods are: using the Apriori algorithm to mine the individual movement patterns of each user, to find out the internal factors that affect the user's check-in; using the dynamic time warping algorithm DTW to calculate the similarity between the individual movement patterns of the users; clustering the individual movement patterns of the users Group the modes, get the central mode of each group, and find the external factors that affect the check-in; use the individual movement mode and the overall movement mode to train the Markov model; train the Markov chain model based on IMP and AMP to predict the user's Next position; consider the influence of external weather, create general weather characteristics; use Gaussian kernel function to calculate the similarity between the weather at the current location and the weather at other locations, and revise the predicted results; set evaluation standards and benchmark methods. The present invention makes the predicted results more realistic.

Description

A Location Prediction Method Based on User Movement Patterns

technical field

The invention belongs to the field of machine learning, and relates to a position prediction method based on user movement patterns.

Background technique

With the popularization of mobile terminals, it is easier to obtain human movement data. Location-based social networking platforms have also collected a large amount of user check-in data. The study of human movement patterns has become a hot topic, and it has become possible to study people's movement patterns. Among them, location prediction is more common. Through location prediction, it is possible to know the user's mobile preference in advance, as well as the movement tendency of the flow of people, which can not only provide targeted services to users, but also bring benefits to businesses. The existing research mainly analyzes the user's behavior through the user's check-in history, finds the user's movement pattern, and then predicts the location. Among them, most of the factors considered are time, space, social, etc., which mainly focus on the user's preference, ignoring the connection between locations. In addition, most studies use the user's personal movement patterns to predict, if the user goes to a place that has never been before, there is no data available to train the model; some researchers use the overall check-in data to train the model, so that the model It can be applied to the location prediction of all users. However, the prediction based on the overall data is too coarse-grained. If the users are currently in the same place, the final prediction results are all the same place, which is inconsistent with the actual situation.

Aiming at the problem that the traditional position prediction model based on discrete state sequence cannot predict the position well, the present invention considers the correlation between different positions in the user's check-in track, and mines the user's individual movement pattern and the overall user's movement pattern from the user's historical check-in data. mobile mode, that is, the internal and external factors that affect user check-in. Among them, the overall movement pattern is mainly reflected by the group's behavior; the individual movement pattern considers the dynamic changes of the user's personal movement pattern and individualized location prediction. The time factor is also taken into account in the process of mining the user's movement pattern. According to the change of the user's movement pattern in the two different time periods of weekdays and weekends, the user's movement pattern in different time periods of the week is respectively mined, which can not only reflect The actual transition between user space locations also implies the underlying temporal regularity of their movement, thus making the predicted results more realistic. In addition, an overall weather feature was created, and a Gaussian kernel function was used to calculate the weather similarity to correct the results.

SUMMARY OF THE INVENTION

In view of this, an object of the present invention is to provide a method for location prediction based on user movement patterns.

To achieve the above object, the present invention provides the following technical solutions:

A method for location prediction based on user movement patterns, the method includes the following steps:

The mobile mode MP is defined as: the set of locations that the user sequentially visits in continuous time; the mode in which the user moves frequently is called the user's mobile mode, and the mobile mode is expressed as MP={l ₁ ,l ₂ ,l ₃ ,...,l _n }, n is the number of positions included in the user's mobile mode;

The individual movement pattern IMP is defined as: the frequently-occurring position sequence in the user's personal historical access location, for a given user, its movement pattern is the set of all movement patterns in the historical check-in record;

The support degree is defined as: the frequency of the user's movement pattern appearing in his movement trajectory; in the user's historical check-in record, which contains multiple movement trajectories, the support degree of the movement pattern can be calculated as:

The overall mobility pattern AMP is defined as: the mobility patterns that frequently appear in the historical check-ins of all users in different groups; the similarity of user mobility patterns is calculated by using the DTW algorithm, and then divided into multiple groups by clustering, according to all users in each group. The movement patterns in the historical access locations find a central pattern, which is the overall movement pattern of all users in this group;

Define the general weather characteristics: combine the three weather characteristics of rainfall, temperature and wind speed into a new feature according to the method of weighted fusion;

S1: Use the Apriori algorithm to dig out the individual movement patterns of each user, and find out the internal factors that affect the user's check-in;

S2: Use the dynamic time warping algorithm DTW to calculate the similarity between the individual movement patterns of users;

S3: Group the individual movement patterns of users through clustering to obtain the central pattern of each group, that is, the overall movement pattern AMP, and find the external factors that affect check-in;

S4: Use the individual movement patterns and the overall movement patterns to train the Markov model respectively;

S5: Train the Markov chain model based on IMP and AMP, combine the probability vectors of the two, and predict the user's next location;

S6: Consider the influence of external weather, and create general weather characteristics;

S7: Use the Gaussian kernel function to calculate the similarity between the weather at the current location and the weather at other locations, and correct the predicted result;

S8: Set evaluation criteria and benchmark methods.

Optionally, the step S1 is specifically:

S11: In a given time range, by analyzing Gowalla, find out the movement pattern of length 1 in the user's check-in record;

S12: Next, find out the movement patterns with a length of 2 in turn, and then calculate whether the support degree σ meets the requirements, and keep repeating this cycle until the length of the movement patterns cannot be increased, and a candidate movement pattern is obtained;

S13: From the obtained candidate movement patterns, find out the movement patterns whose support degree satisfies the condition, and obtain the individual movement patterns of the user.

Optionally, in the step, for the similarity calculation of the two movement modes, the Euclidean distance between the two points is not simply calculated, but the Haversine distance is calculated, and the coordinates of the two points are passed in to obtain the actual geographic space between the two points. distance, as follows:

in:

|M _p | represents the length of the moving pattern, that is, the number of positions in the pattern; rest(M _p ) represents the moving pattern that removes the first location, and d(l,l _i ) represents the true distance between the two locations.

Optionally, the step S3 is specifically:

S31: Initialize multiple classes according to the user's personal movement pattern, and set a distance threshold τ;

S32: For each movement pattern of each user, calculate its distance from each class, and select the class with the smallest distance;

S33: Then use the DTW algorithm to calculate the distance between the movement pattern and this class, if it is less than the threshold τ, add it to the class and update it; otherwise, create a new class for the movement pattern;

S34: Obtain the result of clustering, that is, the overall movement pattern to which each person belongs.

Optionally, the step S4 is specifically:

S41: After clustering the personal movement patterns of the users, obtain the overall movement patterns of the users, and combining the obtained overall movement patterns, the next location to be visited is:

S42: Based on the personal movement pattern, the next location to go is:

表示含有N个位置的移动模式，MP_c表示移动模式类的集合，

表示序列为

的移动模式出现在MP_c中的次数，

表示在MP_c中，位置l_i紧随后面出现的次数。in,

represents a mobility pattern with N positions, MP _c represents a collection of mobility pattern classes,

represents the sequence as

The number of times that the movement pattern appears in MP _c ,

represents the number of occurrences of position _li immediately following in MP _c .

Optionally, for each person in the step S5, there is a personal movement pattern and an overall movement pattern, which are respectively used for the training of the Markov model; finally, a predicted probability vector will be obtained; the overall vector is P ^AMP =(l ₁ ,l ₂ ,l ₃ ,...,l _n ), the vector based on individual movement patterns is P ^IMP =(l ₁ ,l ₂ ,l ₃ ,...,l _n ), where n represents the predicted position the number of ; then combine the two obtained results to obtain the final prediction result; the final combination is as follows:

P=α·P ^IMP +(1−α)P ^AMP .

Optionally, the step S6 includes:

S61: Create a general weather feature X _weather = [Temperature, Rain, Windspeed];

S62: Perform a weighted summation of the three weathers at the user's check-in location, and comprehensively consider the effects of the three weathers on the user's check-in location to obtain the general weather characteristics of each user's check-in location, which are specifically expressed as follows:

X _weather = ω ₁ Temperature+ω ₂ Windspeed+ω ₃ Rain

Among them, the weight of rainfall is calculated as follows:

指的是用户签到的其中一个地点l_i在给定降雨区间用户签到的总次数，

相应时间段内该降雨区间出现的总天数；风速和温度的权重计算也一致。

refers to the total number of user check-ins in one of the locations l _i checked in by the user in a given rainfall interval,

The total number of days that the rainfall interval occurs in the corresponding time period; the weight calculation of wind speed and temperature is also the same.

天气的相似度，得到最终的预测结果；具体计算如下：Optionally, in the step S7, after calculating the user's weather preference through the created weather total feature, a Gaussian kernel function is used to calculate the user's current location _X1 and other locations.

The similarity of the weather is used to obtain the final prediction result; the specific calculation is as follows:

则为其他地点的天气情况。Among them, X _l represents the weather condition of the user's current location,

The weather conditions at other locations.

Optionally, the step S8 includes:

S81: Accuracy and APR of location prediction are used as evaluation criteria for the experiment;

S82: Accuracy: This indicator defines the proportion of correctly predicted locations in the total predicted locations in the user's prediction result list; p(l)=1 when the prediction result is consistent with the actual;

S83: Average percentage ranking APR: The prediction problem also has a certain relationship with the ranking. The check-in location l _j of user _ui is defined in the prediction list PR as:

Take the average of the sum of PR values to get the APR values of all users. The larger the value, the better the prediction effect; the formula is as follows:

S84: In order to verify the effectiveness of the proposed method for location prediction based on user movement patterns, the following models are selected for comparison with the proposed model:

NextPlace: It is a classic location prediction method that predicts user behavior based on nonlinear time series analysis of arrival time, and uses the similarity of time series to make predictions;

SimPreT: Correlate historical patterns with current user trajectories and use pattern similarity to determine the user's next location;

HMM-based: This model builds a mixed Markov model while taking into account non-Gaussian as well as spatiotemporal properties in actual human check-in data.

The beneficial effect of the present invention is that: the present invention considers the association between different positions in the user's check-in track, and respectively excavates the user's individual movement pattern and the overall movement pattern from the user's check-in history data, that is, the internal and external factors that affect the user's check-in. . Among them, the overall movement pattern is mainly reflected by the group's behavior; the individual movement pattern considers the dynamic changes of the user's personal movement pattern and individualized location prediction. The time factor is also taken into account in the process of mining the user's movement pattern. According to the change of the user's movement pattern in the two different time periods of weekdays and weekends, the user's movement pattern in different time periods of the week is respectively mined, which can not only reflect The actual transition between user space locations also implies the underlying temporal regularity of their movement, thus making the predicted results more realistic. In addition, an overall weather feature was created, and a Gaussian kernel function was used to calculate the weather similarity to correct the results.

Other advantages, objects, and features of the present invention will be set forth in the description that follows, and will be apparent to those skilled in the art based on a study of the following, to the extent that is taught in the practice of the present invention. The objectives and other advantages of the present invention may be realized and attained by the following description.

Description of drawings

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be preferably described in detail below with reference to the accompanying drawings, wherein:

Fig. 1 is the overall flow chart of the present invention;

Figure 2 is the check-in location information in the dataset;

Figure 3 is a comparison of the prediction accuracy of different movement modes;

Figure 4 shows the comparison of the accuracy of the model in the two cities;

Figure 5 shows the comparison of model APR values.

Detailed ways

The embodiments of the present invention are described below through specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments are only used to illustrate the basic idea of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

Among them, the accompanying drawings are only used for exemplary description, and represent only schematic diagrams, not physical drawings, and should not be construed as limitations of the present invention; in order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings will be omitted, The enlargement or reduction does not represent the size of the actual product; it is understandable to those skilled in the art that some well-known structures and their descriptions in the accompanying drawings may be omitted.

The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms “upper”, “lower”, “left” and “right” , "front", "rear" and other indicated orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the indicated device or element must be It has a specific orientation, is constructed and operated in a specific orientation, so the terms describing the positional relationship in the accompanying drawings are only used for exemplary illustration, and should not be construed as a limitation of the present invention. situation to understand the specific meaning of the above terms.

The present invention is a position prediction method based on the user's movement pattern. The Markov model is trained separately through the excavated individual and overall movement patterns, so as to predict the user's location, and propose a location based on the similarity between the user's individual and the overall movement pattern. predict. First, the movement patterns of each user are mined; then the DTW algorithm is used to cluster according to the similarity, and the overall movement patterns are obtained; then the Markov models are trained with the individual movement patterns and the overall movement patterns to predict the location. . Finally, an overall weather feature is also created, and a Gaussian kernel function is used to calculate the weather similarity to correct the results.

For a more concise and clear description, some definitions of terms are explained:

A mobility pattern (MP) is defined as: a set of locations that a user sequentially visits in continuous time. The frequent movement pattern of the user is called the user's movement pattern, and the movement pattern can be expressed as MP={l ₁ , l ₂ , l ₃ , . . . , l _n }, where n is the number of positions included in the user's movement pattern.

The individual movement pattern (IMP) is defined as: the frequently-occurring location sequence in the user's personal historical visit location, for a given user, its movement pattern is the set of all movement patterns in the historical check-in records.

The support degree is defined as: the frequency of the user's movement pattern in his movement trajectory. In the user's historical check-in record, including multiple movement trajectories, the support degree of the movement mode can be calculated as:

The overall mobility pattern (AMP) is defined as: the mobility pattern that frequently appears in the historical check-ins of all users in different groups. The DTW algorithm is used to calculate the similarity of user movement patterns, and then it is divided into multiple groups by clustering, and a central pattern is found according to the movement patterns in the historical access locations of all users in each group, which is the overall movement of all users in this group. model.

Define the general weather characteristics: combine the three weather characteristics of rainfall, temperature, and wind speed into a new feature according to the weighted fusion method.

As shown in Figure 1, the present invention is divided into the following steps:

S1: Use the Apriori algorithm to mine the individual movement patterns of each user, and find out the internal factors that affect the user's check-in.

S2: Calculate the similarity between the individual movement patterns of the users by using the Dynamic Time Warping (DTW).

S3: Group the individual movement patterns of users through clustering, obtain the central pattern of each group, that is, the overall movement pattern (AMP), and find the external factors that affect the check-in.

S4: Train the Markov model with the individual movement patterns and the overall movement patterns, respectively.

S5: Train a Markov chain model based on IMP and AMP, combine the probability vectors of the two, and predict the user's next location.

S6: Consider the influence of external weather, and create a general weather feature.

S7: Use the Gaussian kernel function to calculate the similarity between the weather at the current location and the weather at other locations, and correct the predicted result.

S8: Set evaluation criteria and benchmark methods.

Among them, in step S1, the Apriori algorithm is used to mine the user's movement pattern, and the Apriori algorithm is a data mining algorithm based on association rules, which is later applied to the mining of the user's movement pattern. We added the time factor in the mining process of mobile patterns, so that the excavated user mobile patterns have time regularity, and we can know the changes of users' mobile patterns with time. First, in a given time range (weekdays, weekends), through the analysis of Gowalla, find out the mobile pattern of length 1 in the user's check-in record. The check-in information is shown in Figure 2. Then find out the movement patterns with length 2 in turn, and then calculate whether its support σ meets the requirements, and keep repeating this cycle until the length of the movement patterns cannot be increased, and the candidate movement patterns are obtained in this way. From the obtained candidate movement patterns, find out the movement patterns whose support degree satisfies the condition, and obtain the user's individual movement pattern.

Among them, in step S2, for the similarity calculation of the two moving modes, we do not simply calculate the Euclidean distance between the two points, but the calculated Haversine distance, and the actual distance between the two points can be obtained by passing in the coordinates of the two points. Geospatial distance, so the calculated distance is more accurate. details as follows:

in:

|M _p | indicates the length of the movement pattern, that is, the number of positions in the pattern; rest(M _p ) indicates the movement pattern with the first position removed, and d(l,l _i ) indicates the true distance between the two positions .

Wherein, in step S3, the user's personal movement pattern is obtained according to step S1, a plurality of classes are initialized, and a distance threshold τ is set. For each movement pattern of each user, calculate its distance to each class, and select the class with the smallest distance. Then the DTW algorithm is used to calculate the distance between the movement pattern and this class, if it is less than the threshold τ, it is added to the class and updated, otherwise, a new class is created for the movement pattern. Finally, the result of clustering is obtained, that is, the overall movement pattern to which each person belongs.

Among them, in step S4, after clustering the user's personal movement pattern according to step S3, we obtain the overall movement pattern of the user, and combined with the obtained overall movement pattern, the next position he will go to is:

And based on the personal movement pattern, the next location it will go to is:

表示含有N个位置的移动模式，MP_c表示移动模式类的集合，

表示序列为

的移动模式出现在MP_c中的次数，

表示在MP_c中，位置l_i紧随后面出现的次数。in,

represents a mobility pattern with N positions, MP _c represents a collection of mobility pattern classes,

represents the sequence as

The number of times that the movement pattern appears in MP _c ,

represents the number of occurrences of position _li immediately following in MP _c .

Wherein, in step S5, for each person, there is a personal movement pattern and an overall movement pattern, which are respectively used for the training of the Markov model. In the end, a predicted probability vector will be obtained. The ensemble-based vector is P ^AMP =(l ₁ ,l ₂ ,l ₃ ,...,l _n ), the individual movement pattern-based vector is P ^IMP =(l ₁ ,l ₂ ,l ₃ ,...,l _n ), where n represents the number of predicted positions; then the two obtained results are combined to obtain the final predicted result. So the final combination is as follows:

P=α·P ^IMP +(1-α)P ^AMP

Wherein, in step S6, firstly, the total weather feature X _weather =[Temperature, Rain, Windspeed] is created. Then, the weighted summation of the three weathers at the user's check-in location is carried out, and the influence of the three weathers on the user's check-in location is comprehensively considered, and the general weather characteristics of each user's check-in location are obtained, which are specifically expressed as follows:

X _weather = ω ₁ Temperature+ω ₂ Windspeed+ω ₃ Rain

Among them, the weight of rainfall is calculated as follows:

指的是用户签到的其中一个地点l_i在给定降雨区间用户签到的总次数，

相应时间段内该降雨区间出现的总天数。风速和温度的权重也类似的计算。

refers to the total number of user check-ins in one of the locations l _i checked in by the user in a given rainfall interval,

The total number of days that this rainfall interval occurs during the corresponding time period. The weights for wind speed and temperature are calculated similarly.

天气的相似度，得到最终的预测结果。具体计算如下：Among them, in step S7, after calculating the user's weather preference through the created general weather features according to S6, we use a Gaussian kernel function to calculate the user's current location (X _l ) and other locations

Similarity of weather to get the final prediction result. The specific calculation is as follows:

则为其他地点的天气情况。Among them, X _l represents the weather condition of the user's current location,

The weather conditions at other locations.

Among them, in step S8, the check-in records of two cities in the Gowalla data set for one year are selected as the data of London (LON) and Los Angeles (LA) as the test data set. The data of the experiment is divided into test set, training set and validation set. After training on the training set, it is used for verification on the validation set, and finally tested on the test set. The information of the dataset is shown in Figure 2:

Accuracy and APR of location prediction were used as evaluation criteria for the experiment. Defined as follows:

Accuracy: This metric defines the proportion of correctly predicted locations to the total predicted locations in the user's prediction result list. p(l)=1 when the predicted result is consistent with the actual.

Average Percentile Rank (APR): The prediction problem also has a certain relationship with the ranking. The check-in location l _j of user _ui is defined in the prediction list PR as:

The APR value of all users is obtained by taking the average of the sum of the PR values. The larger the value, the better the prediction effect. The formula is as follows:

Secondly, in order to verify the effectiveness of the proposed method for location prediction based on user movement patterns, the following models are selected for comparison with the proposed model:

NextPlace: It is a classic location prediction method that predicts user behavior based on non-linear time series analysis of arrival time, and uses the similarity of time series to make predictions.

SimPreT: Correlate historical patterns with current user trajectories and leverage pattern similarity to determine the user's next location.

HMM-based (Hybrid Markov Model-based): This model constructs a hybrid Markov model while taking into account the non-Gaussian and spatiotemporal characteristics of actual human check-in data.

Figure 3 shows the prediction performance of individual mobility patterns (IMP) and overall mobility patterns (AMP). From the figure, we can see that when only the overall movement pattern is considered, the prediction performance gradually decreases with the increase of the location; the method based only on the individual movement pattern obviously encounters the cold start problem in the early stage, and the prediction accuracy is low. However, with the accumulation of individual historical activity information, the prediction accuracy is greatly improved. Especially in late stages, it outperforms methods based on individual movement patterns and overall activity patterns. Nonetheless, in our approach we still consider both individual and overall mobility patterns. We believe that preserving the overall patterns of users makes our prediction method more robust and can handle some cases that cannot be predicted by individual movement patterns alone.

The comparison of the accuracy of the model proposed by the present invention and other models is shown in FIG. 4 . It can be seen from the figure that the accuracy of the model proposed by the present invention is much higher than that of the NextPlace and SimPreT algorithms, and slightly higher than that of HMM-based. In the LA dataset, the improvements were increased by 15%, 5%, and 2.1%, respectively; in the LON dataset, they were increased by 14%, 6.5%, and 4.2%, respectively. Explain that taking into account both the user's individual and overall movement patterns, as well as taking into account the weather, can help improve forecast results.

As shown in Figure 5, in LA, the APR performance of the proposed model is better than other models, the APR of the model is 19% higher than that of NextPlace, 7% higher than that of SimPreT model, and higher than that of HMM-based model. At the same time, on the LON dataset, the model of the present invention is 18% higher than NextPlace, 9.2% higher than SimPreT model, and 5% higher than HMM-based model.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be Modifications or equivalent replacements, without departing from the spirit and scope of the technical solution, should all be included in the scope of the claims of the present invention.

Claims (9)

1. a position prediction method based on user movement pattern is characterized in that: the method comprises the following steps: The mobile mode MP is defined as: the set of locations that the user sequentially visits in continuous time; the mode in which the user moves frequently is called the user's mobile mode, and the mobile mode is expressed as MP={l ₁ ,l ₂ ,l ₃ ,...,l _n }, n is the number of positions included in the user's mobile mode; The individual movement pattern IMP is defined as: the frequently-occurring position sequence in the user's personal historical access location, for a given user, its movement pattern is the set of all movement patterns in the historical check-in record; The support degree is defined as: the frequency of the user's movement pattern appearing in his movement trajectory; in the user's historical check-in record, which contains multiple movement trajectories, the support degree of the movement pattern can be calculated as:

The overall mobility pattern AMP is defined as: the mobility patterns that frequently appear in the historical check-ins of all users in different groups; the similarity of user mobility patterns is calculated by using the DTW algorithm, and then divided into multiple groups by clustering, according to all users in each group. The movement patterns in the historical access locations find a central pattern, which is the overall movement pattern of all users in this group; Define the general weather characteristics: combine the three weather characteristics of rainfall, temperature and wind speed into a new feature according to the method of weighted fusion; S1: Use the Apriori algorithm to dig out the individual movement patterns of each user, and find out the internal factors that affect the user's check-in; S2: Use the dynamic time warping algorithm DTW to calculate the similarity between the individual movement patterns of users; S3: Group the individual movement patterns of users through clustering to obtain the central pattern of each group, that is, the overall movement pattern AMP, and find the external factors that affect check-in; S4: Use the individual movement patterns and the overall movement patterns to train the Markov model respectively; S5: Train the Markov chain model based on IMP and AMP, combine the probability vectors of the two, and predict the user's next location; S6: Consider the influence of external weather, and create general weather characteristics; S7: Use the Gaussian kernel function to calculate the similarity between the weather at the current location and the weather at other locations, and correct the predicted result; S8: Set evaluation criteria and benchmark methods. 2. a kind of position prediction method based on user movement pattern according to claim 1, is characterized in that: described step S1 is specifically: S11: In a given time range, by analyzing Gowalla, find out the movement pattern of length 1 in the user's check-in record; S12: Next, find out the movement patterns with a length of 2 in turn, and then calculate whether the support degree σ meets the requirements, and keep repeating this cycle until the length of the movement patterns cannot be increased, and obtain the candidate movement patterns; S13: From the obtained candidate movement patterns, find out the movement patterns whose support degree satisfies the condition, and obtain the individual movement patterns of the user. 3. a kind of position prediction method based on user's movement pattern according to claim 2, is characterized in that: in described step, for the similarity calculation of two kinds of movement patterns, it is not simple to calculate the Euclidean distance between two points, To calculate the Haversine distance, pass in the coordinates of two points to get the actual geographic space distance between the two points, as follows:

in:

|M _p | represents the length of the movement pattern, that is, the number of positions in the pattern; rest(M _p ) represents the movement pattern with the first position removed, and d(l,l _i ) represents the true distance between the two positions. 4. a kind of position prediction method based on user movement pattern according to claim 3, is characterized in that: described step S3 is specifically: S31: Initialize multiple classes according to the user's personal movement pattern, and set a distance threshold τ; S32: For each movement pattern of each user, calculate its distance from each class, and select the class with the smallest distance; S33: Then use the DTW algorithm to calculate the distance between the movement pattern and this class, if it is less than the threshold τ, add it to the class and update it; otherwise, create a new class for the movement pattern; S34: Obtain the result of clustering, that is, the overall movement pattern to which each person belongs. 5. a kind of position prediction method based on user movement pattern according to claim 4, is characterized in that: described step S4 is specifically: S41: After clustering the personal movement patterns of the users, obtain the overall movement patterns of the users, and combining the obtained overall movement patterns, the next location to be visited is:

S42: Based on the personal movement pattern, the next location to go is:

in,

represents a mobility pattern with N positions, MP _c represents a collection of mobility pattern classes,

represents the sequence as

The number of times that the movement pattern appears in MP _c ,

represents the number of occurrences of position _li immediately following in MP _c . 6. a kind of position prediction method based on user's movement pattern according to claim 5, is characterized in that: described step S5 has its personal movement pattern and overall movement pattern for each person, and it is used for Maldives respectively. Training of the Kov model; a predicted probability vector will be obtained in the end; the vector based on the whole is P ^AMP = (l ₁ ,l ₂ ,l ₃ ,...,l _n ), and the vector based on the individual movement pattern is P ^IMP =( l ₁ ,l ₂ ,l ₃ ,…,l _n ), where n represents the number of prediction positions; then combine the two obtained results to obtain the final prediction result; the final combination is as follows: P=α·P ^IMP +(1−α)P ^AMP . 7. The method for position prediction based on user movement patterns according to claim 6, wherein the step S6 comprises: S61: Create a general weather feature X _weather = [Temperature, Rain, Windspeed]; S62: Perform a weighted summation of the three weathers at the user's check-in location, and comprehensively consider the effects of the three weathers on the user's check-in location, to obtain the general weather characteristics of each user's check-in location, which are specifically expressed as follows: X _weather = ω ₁ Temperature+ω ₂ Windspeed+ω ₃ Rain Among them, the weight of rainfall is calculated as follows:

refers to the total number of user check-ins in one of the locations l _i checked in by the user in a given rainfall interval,

The total number of days that the rainfall interval occurs in the corresponding time period; the weight calculation of wind speed and temperature is also the same. 8. A kind of position prediction method based on user's movement pattern according to claim 7, it is characterized in that: in described step S7, after the weather preference of user is calculated by the weather general feature created, adopt Gaussian kernel function to calculate User's current location X _l and other locations

The similarity of the weather is used to obtain the final prediction result; the specific calculation is as follows:

Among them, X _l represents the weather condition of the user's current location,

The weather conditions at other locations. 9. The method for position prediction based on user movement patterns according to claim 8, wherein the step S8 comprises: S81: Accuracy and APR of location prediction are used as evaluation criteria for the experiment; S82: Accuracy: This indicator defines the proportion of correctly predicted locations in the total predicted locations in the user's prediction result list; p(l)=1 when the prediction result is consistent with the actual;

S83: Average percentage ranking APR: The prediction problem also has a certain relationship with the ranking. The check-in location l _j of user _ui is defined in the prediction list PR as:

Take the average of the sum of PR values to get the APR values of all users. The larger the value, the better the prediction effect; the formula is as follows:

S84: In order to verify the effectiveness of the proposed method for location prediction based on user movement patterns, the following models are selected for comparison with the proposed model: NextPlace: It is a classic location prediction method that predicts user behavior based on nonlinear time series analysis of arrival time, and uses the similarity of time series to make predictions; SimPreT: Correlate historical patterns with current user trajectories, and use pattern similarity to determine the user's next location; HMM-based: This model builds a mixed Markov model while taking into account the non-Gaussian as well as spatiotemporal properties of actual human check-in data.

Images