Keywords

1 Introduction

This chapter is aimed to provide researchers a review on recent techniques and methods used for Location recommendation system based on LBSNs. The point of becoming a smart city is that it will increase resilience and improve the lives of citizens. There are various developments taking place for incorporating smart cities [1]. Various trends and technologies are emerging for various applications such as healthcare [2], transport [3], home appliances etc. The vision of a smart city is to implement more technology or to explore how technology might enable the city and citizens to solve the challenges they face. Here with the collaboration of Location-based Social Network (LBSN) in the smart cities which recommends the citizens by Point-of-Interest methodology makes it easier for people to visit different people according to their preferences easily [4, 5]. An expanding advancement and notoriety of Location-based social networks (LBSNs) has driven to creation on large-scale check-in dataset. Recent researches on the recommender systems show that the usage of social network data could provide quite improved and more personalized recommendations with the better accuracy of prediction [24]. Due to lack of knowledge and awareness about the new sites, the individuals tend to visit only traditional and usual places. But due to the advancement in the social media the users can express and share their experiences online to the web communities which help in analyzing their taste and preferences online and suggest places to them and their friends online. Progressively numerous clients post their whereabouts, visiting hour, ratings, etc. in the type of registration records and offer their experiences. Contrarily, when clients face enormous measures of data on LBSNs, recommender frameworks endeavor to suggest the most appropriate things (e.g., areas, companions, music and promotions) to clients by using the colossal client registration information asset, which can mitigate the issue of data overload. The propelled data advances that have come about because of the quick development of Location based administrations have incredibly improved individuals’ urban lives [25].

This chapter deals with the review of various technologies used for online location recommendation systems for smart cities. Further the technology proposed in this chapter provides us useful insights for developing personalized POI recommendations for the smart communities using various features. Figure 1 shows a general LBSN model. LBSNs enable users to register and provide their regions, experiences, and encounters about focal points called the Point-of-interests (POIs) with their companions whenever and anyplace. For instance, while eating at a café, we may capture photographs of the food and quickly share these photographs with our companions by means of LBSNs. These registration conducts make it real day by day encounters disperse rapidly over the cyberspace. Also, the registration information can be completely be misused to comprehend the inherent regulations of day to day human progress and portability, that can be applied for recommender frameworks and field-based administration. In this manner, field-based web-based life information administration are pulling in the significant consideration from various business areas, for instance, client profiling, suggestion frameworks, urban crisis occasion the board, urban arranging, and advertising choices. The client create spatial-transient information that could be gathered from the LBSNs and could be commonly used for comprehension as well as for displaying human portability as per the accompanying four viewpoints [6, 7].

Topographical Feature: The spatial highlights of human development as covered up in a huge number of registrations, information has been abused to comprehend human portability. For instance, individuals will in general move to close by spots and every so often too inaccessible spots: the previous is short-gone travel and isn’t influenced by informal community ties that are occasional both geographically and transiently. The final one is long-separation travel and more affected by interpersonal organization connections.

Temporal Features: With respect to the schedule of our day to day lives, there are various probabilities for us to visit various locales at various hours of the day and various days of the week. The registration information of LBSBs similarly uncovers such outcomes. A great many people get down to business on the weekdays, their registration practices frequently occur around early afternoon or around evening time, and the areas they pick are near their working environments or homes. On the weekends, most of the registrations occur at the beginning of the day or evening, and the areas are near sure POIs (e.g., a commercial center, eatery, gallery, or beautiful spot).

Fig. 1
figure 1

A general model for LBSNs

Full size image

Social Features: To begin with, many researches contemplates to show that individuals will, in general, visit places that are nearest to their place more frequently than the spots that are far away, however they will in general visit far off spots near their companions’ place and also the places which are visited by their companions. Such perceptions are broadly utilized for area proposals. Secondly, the spatio-temporal element abstracted from the information in registration is exploited in order to construe social connections and companion proposals.

Incorporated Feature: The registration information in LBSNs, provides an approach to discover the user’s geographical and ephemeral impulse, and collect their social connections. In addition, it generally gives another point of view from which the related financial exhibitions and urban structures could be depicted, road systems and point-of-interest ubiquity could be evaluated, intra-urban development streams can be examined in urban territories, urban major/crisis occasions can be distinguished and financial effects of social ventures can be identified. Despite the fact that few overviews on POI proposals have been distributed, hardly any examinations hand out interpretations of client displaying for venue suggestions as well as group existing client demonstrating different approaches dependent on the kind of LBSNs information. The paper centers around surveying how we can effectively utilize client produced information to show POI suggestions. The commitments of this paper are as per the following:

  1. (1)

    Presents the framework and information qualities of LBSNs in a nutshell.

  2. (2)

    Taking the qualities of topographical as well as the social information in LBSNs into consideration, we present an interpretation of client demonstrating POI proposals.

  3. (3)

    According to the kind of LBSN information that is completely used in client demonstrating approaches for POI proposals, the client displaying calculations can be isolated into four categories: unadulterated registration information based client demonstrating, geological data based client displaying, spatio-worldly data-based client demonstrating, and geo social data-based client demonstrating.

Specifically, it is significant and helpful to make proposals when a client visits a new territory; in this way, POI suggestion has been brought into the LBSNs administrations. POI suggestion could prescribe spots for the clients where they have not been before by extracting clients’ inclinations and interests dependent on the LBSNs authentic records that have significant down to earth hugeness and hypothetical worth. There are a few novel attributes of LBSNs which recognize POI proposals from conventional suggestion undertakings.

  • The Tobler’ law states that “Everything is identified with everything else, except close to things are more related than the inaccessible thing”. This law demonstrates that the topographically close POIs are bound to have comparative qualities. Likewise, the likelihood of POI chosen for the client is conversely relative to the geographical separation [8].

  • Regional fame: Two point-of-interests with comparable or equivalent semantic points can end up having different popularities in the event where they are situated in distinct locales.

  • Dynamic client versatility: In a location-based social networks, a client may visit POIs at various locales, for instance, a client may make a trip to various urban communities. Dynamic client portability forces gigantic difficulties on POI proposals.

  • Implicit client input: In the investigation of POI proposals, the unequivocal client evaluations are normally not accessible. The recommender framework needs to construe client inclinations from understood client criticism as far as client registration recurrence information.

Clients’ exercises are frequently influenced by time. For instance, a client is bound to go to a café for lunch as opposed to a bar, around early afternoon and is bound to go to a bar instead of a Mall at 12 PM. Along these lines, the proposal results ought to be time mindful or time-arranged. Be that as it may, as far as we could possibly know, none of the current approaches have taken the temporal factor into consideration for POI proposals in LBSNs. Likewise there are some approaches which consider the time factor but they do not consider the other factors such as social influence, geographical influence, etc. [7]. It is indeed a great challenge to suggest a locale to the user based on the time he is visiting that places as well as the popularity of the place at that time, and the place which is nearest to them. So in this chapter, we discuss about various methods and approaches used during the course of time for recommending the accurate POI for the users considering various factors that influences the user in an LBSN to visit a particular place [9]. We also propose architecture for location recommendation which integrates the spatial factor with the user social and activity feature. The major contributions of this chapter are as follows:

  • It reviews various features and techniques required for the Point-of-interest recommendation system based on the location based web communities.

  • It discusses the work that has been done on POI recommendation system for online communities.

  • It also proposes a Location recommendation algorithm for smart cities using Foursquare dataset for experimenting the results.

The chapter is organized as follows. Section 2 discusses various POI based recommendation systems based on topographical features. In Sect. 3 we discuss various POI based recommendation systems based on temporal features. In Sect. 4 we discuss the various techniques used for POI based recommendation systems based on user behavior. Section 5 discusses the POI based recommendation systems based on integration of various features. In Sect. 6 we discuss the proposed work, methodology used and the experimental results achieved using Foursquare dataset. In Sect. 7 we talk about the conclusion and future scope for location recommendation systems based on our observation.

2 POI Based Recommendation Systems Based on Topographical Features

There are various techniques and methodologies used to suggest the locale based on the POIs for LSBNs based on the topographical features. Some of them are explained below.

2.1 Mining Topographical Impact for Collaborative POI Recommendation

In this method, they have focused on the implementation of a POI based recommender service for the LBSNs. They have considered three parameters for the recommendation which include user preference, topographical features and social impact from the user’s friends. Though they have considered the three parameters as mentioned they have focused more on the geographical impact. The reason behind this is the presence of the spatial clustering aspect present in the user check-in activities. This is implemented by using power-law distribution. They have used the naive Bayesian method to implement a collaborative recommendation method [10]. Here a consolidated POI recommendation scheme is used which unifies the POI with social influence according to the user’s preference and the venue’s geographical features. This method has experimented Foursquare and Whrrl. The outcome with these datasets proved to be much better than the other alternative recommendation strategies.

2.2 Exploring Geographical Inclinations for POI Recommendation

POI recommendations mostly prefer to provide personalized recommendation and they are often quite complex. They depend on various factors such as spatial features, user behavior, etc. Most of the POI recommendations used to lack the integrated investigation of the joint effect of various aspects. In this method, a spatial probabilistic factor analysis system is proposed. This framework strategically takes various factors into consideration. It allows securing the spatial influence along with the check-in behavior of the client. It also integrates the portability behavior of the users which plays a major role in recommending places according to the user preferences [11]. It also considers the check-in feature of the user as feedback from the user to understand the user’s preferences. To illustrate POI over an examined area a Gaussian distribution was utilized. This is based on the law of geography which states that similar POIs are more akin instead of the POI which is far away from each other. In order to implement the mobility behavior of the user over various activity regions a multinomial distribution was used over the latent region. This method has experimented on Foursquare dataset. This method proves to outperform the rest of the newfangled, latent factor models with a substantial margin.

2.3 Integrating Matrix Factorization with Joint Geographical Modeling (GeoMF) Method for POI Recommender System

POI recommendation helps the user to explore and discover new places to visit. For a recommendation of POI based on user preference a user-POI matrix is used. The most critical challenge is to deal with the sparsity of these matrices. To overcome this challenge a model that exploits the weight matrix factorization method is proposed [12]. This provides better collaborative filtering with inherent feedback. In addition, researchers have identified a geographical clustering event in user mobility behavior. Every individual visiting a venue are likely to group or cluster together and have also demonstrated their efficacy POI recommendations. Hence the incorporation of factorization helps to overcome this challenge too. This method augments the person’s as well as the POI’s latent aspects in factorization model the user’s activity and POI’s influence area vectors. The geographical clustering phenomenon is apprehended with the help of two-dimensional kernel density estimation. This methodology clarifies the usage of matrix factorization in the eradication of the matrix sparsity issue. This weighted matrix factorization based model proved to outperform the other factorization methods upon experimentation on a large scale dataset. This model also proved that the integration of the geographical clustering phenomenon with matrix factorization improves the POI recommendation’s performance.

2.4 A Ranking Based Geographical Factorization (Rank-GeoMF) Approach for POI Recommender System

Although sparse check-in data matrix poses a greater problem. It can be overcome by the factorization method. The availability of context data has introduced a new issue about how to utilize them. In this method, a ranking based spatial factorization approach is used which is called Rank-GeoFM based POI recommendation [13]. This method deals with two challenges. The characterization of the preferences of the user is considered with the help of the check-in frequency. This helps to determine the factorization by assigning ranks to the POIs. Here both the POIs without check in and the POIs with check in devotes to the understanding of how the POIs can be ranked and hence the matrix sparsity issue will be able to get resolved. This model also consolidates other contextual information like spatial impact as well as temporal impact. An approach based on the stochastic gradient descent is proposed in order to determine the factorization. In order to examine the efficacy of the methodology that has been proposed both the user-POI as well as the user-time-POI context has experimented. The result of both settings surpasses the other newfangled methods.

2.5 Integration of Geographical Impact with POI Recommender Systems

The choice of recommended POI depends on the client’s choices that are defined based on the following factors: user preferences, social and spatial influence. These factors can be mined from the clients’ check in records. Extracting the client’s preferences and their social influence is quite easier. But mining the geographical influence is a bigger challenge. By employing the Gaussian distribution model we can easily be able to estimate the check-in the behavior of the user. But the results obtained are not satisfactory enough. This method introduces two models. One is the Gaussian mixture model (GMM) and the other is the Gaussian mixture model based on genetic algorithms (GA-GMM) [14]. These both model are used for mining the topographical influences. This method utilizes GMM to automatically determine the activity centers of the user. By eradicating the outliers it exploits GAGMM to enhance GMM. The experimental outcomes using these two methods illustrates that the GMM surpasses the rest of the above discussed PIO recommendation systems that are based on geographical influence, and the GA-GMM removes the effects of the outliers by enhancing the GMM.

2.6 General Topographical Probabilistic Based Factor Approach for Point of Interest Recommendation

POI is based on popular places like Theater, Park, Hotels and etc. Based on the users’ interest and history of users recently visited the place with the present locale of the client it has to decide the probabilistic suggestion on the recommendation to develop the Poisson Geo-PFM to give the better POI with the help of user who recently check-in that location [15]. In a survey of POI recommendations, user ratings are usually not available explicitly. The recommendation system has to interpret client preference from their feedback that is implicit.

2.7 Exploiting Geographical Neighborhood Characteristics for POI Recommender System

As a significant application in LBSNs, the customized area prescribe frameworks (PLRs) can assist clients with investigating new areas to improve their encounters. Then again, PLRs can likewise encourage outsider designers (e.g., publicists) to give increasingly significant administrations in the correct areas. Land attributes got from the authentic registration information have been accounted for compelling in improving area proposal precision. Be that as it may, past examinations basically abuse land attributes from a client’s viewpoint, through displaying the geological conveyance of every individual client’s registration. In this chapter, we are keen on abusing land attributes from an area viewpoint, by displaying the topographical neighborhood of an area. The area is demonstrated at two levels. The first level is example level neighborhood, which is described by a couple of closest neighbors belonging to an area. The second level is the district level neighborhood for the land locale where the area exists. A specific Instance-Region Neighborhood Matrix Factorization (IRenMF) approach has been proposed [6]. This method exploits 2 degrees of topographical neighborhood qualities: (i) the closest neighboring areas will in general offer progressively comparative client inclinations (Occasion level attributes); and (ii) the areas along the equivalent land locale might have comparative client inclinations to area proposal by abusing two degrees of geological neighborhood attributes from area point of view (district level attributes). By joining these two degrees of neighborhood qualities into the learning of dormant variables of clients and areas, IRenMF has a progressively precise displaying of clients’ inclinations on areas. To take care of the advancement issue of this method, a substituting improvement calculation has been proposed, which allows the IRenMF to accomplish stable suggestion exactness. The examinations on genuine information displays that this method prompts considerable enhancements for the old-style MFbased approach and other best in class area suggestion models. In IRenMF, the two degrees of geological attributes are normally fused into the understanding of idle highlights of clients as well as areas, so that it predicts clients’ inclinations on areas all the more precisely. Broad examinations on the genuine information gathered from Gowalla, a well known LBSN, exhibit the viability and points of interest of our methodology.

3 POI Based Recommendation Systems Based on Temporal Features

There are various techniques and methodologies used to suggest the locale based on the POIs for LSBNs based on the temporal features. Some of them are explained below.

3.1 Time-Aware POI Recommendation

There are various recommendation techniques used but many of those do not take the time factor into consideration. Time plays a crucial role since most of people tend to tour various locations depending upon a particular time zone, for instance, restaurants at night and beach in the evening, etc. In this methodology, a time-aware POI recommender system is implemented so as to recommend POI for a given person at a given time zone [16]. Here a collaborative model for recommendation is developed which incorporates the temporal information. Further, the system is enhanced to consider the spatial features also. This methodology has experimented on Foursquare and Gowalla Datasets and this method proved to surpass the newfangled recommendation approaches discussed previously. This method was a pioneer to consider the time factor for POI recommendation. This method unified temporal and spatial features.

3.2 A Probabilistic Framework to Exploit Correlation of Temporal Impact in a Time-Aware Locale Recommender System

Time, significantly influences clients’ registration practices, for instance, individuals for the most part visit better places at various occasions of weekdays and ends of the week, e.g., eateries around early afternoon on weekdays and bars at 12 PM on ends of the week. To make locale suggestion for the clients, most related research infers their choices to POI by exploiting the collaborative filtering techniques with the help of clients’ check in data. Here a probabilistic system to exploit Transient impact relationships for time-aware locale suggestions has been proposed. This is system called TICRec [17]. It beats the two previously mentioned restrictions. The first system is used to dodge the misfortune of temporal data; they have appraised a likelihood thickness on a persistent timeslot of a client going by an area instead of changing the ceaseless time to discrete timeslot openings. The nonstop temporal likelihood densities was demonstrated on the basis of a density estimation strategy which is non parametric, that is, the well-known kernel density estimation also called KDE, as the temporal densities of clients going by areas are exceptionally different and we cannot accept their shapes. The second system was used to assess the time likelihood density of a client going to a locale. They collected (1) a distinctive temporal history of distinctive clients going to an area on the basis of client-based TIC, (2) a distinctive temporal history of a particular client going to diverse areas on the basis of location-based TIC. Appropriately, the temporal history depends on the weekdays and weekends, as their visiting patterns are different. For instance, on weekdays, people mostly attend office while they tend to visit tourist locales on weekends. TICRec conquers two significant constraints in existing time-mindful area suggestion methods. This TICRec model uses KDE technique for assessing a consistent temporal likelihood density for a client who tends to visit another area to maintain a strategic distance from the time data misfortune. To consolidate TIC with TICRec, the time-based likelihood density considers two strategies. The first one is using client-based TIC by corresponding registration practices of various clients for a similar area at various occasions. The second one is locale-based TIC by connecting registration practices of a similar client to various areas at various occasions. The outcome shows that TICRec surpasses the other newfangled time-aware locale recommender systems.

4 POI Based Recommendation Systems Based on User Behavior

There are various techniques and methodologies used to suggest the locale based on the POIs for LSBNs based on the user activity features. Some of them are explained below.

4.1 Exploiting Sequential Influence for Location Recommendation (LORE)

Lots of recommender systems that take spatial, social and temporal features are available. People tend to display a sequential pattern in their movements. This method proposes a new concept called LORE [18] which is intended to employ sequential impact on suggesting the locales. It extracts the patterns that are sequential from the venue and illustrate those patterns in the form of a graph that has dynamic location-location transitions. This graph is called as L2TG. Then it predicts the probability that the person will visit a particular venue by using Additive Markov Chain (AMC). This is used along L2TG. Then it unifies this sequential impact with the spatial and social influence so as to implement a recommendation system [18]. The spatial feature is implemented by using a two-dimensional check-in probability distribution. This method has experimented with Foursquare and Gowalla Datasets and even this approach resulted in better in suggesting the venues than the other newfangled recommender systems that previously discussed.

4.2 Joint Modeling Behavior Based on Check in Approach

For recommending and discovering the interesting and attractive locations based on the Point of Interest by the multi modeling behavior analysis. The wireless communication technologies and location procurement have fostered a number of LBSNs. In LBSNs, it is important to exploit the check-in data to make personalized suggestions. This will help the clients to get familiar with new POIs and discover new locales. In this approach, joint probabilistic generative model (JIM) was proposed to model clients’ check-in behaviors [19]. This model consolidates the factors of content impact, topographical impact, temporal impact as well as the word-of-mouth impact. This makes it effective to overcome the challenges of sparsity of data and client preferences drift across geographical area. To illustrate the application and how flexible JIM is, they have examined about how the out-of-town as well as home-town suggestion schemes are supported in a consolidated manner. They have carried out broad tests in order to assess how the JIM is executed, its efficacy and viability. Then comes the appeared predominance of JIM show over other competitor strategies. Other than, we examined the significance of each figure in making strides both out of town and hometown suggestion beneath one system, and discovered that the substance data performs an overwhelming part to overcome the sparsity of information in out of town recommendation situation, whereas the worldly impact is more critical to make strides hometown suggestion.

4.3 Exploiting User Check-in Data for Location Recommendation in LSBN

In LBSN, clients share data about the areas or spot that they tend to visit along with other information. Visitations are accounted for unequivocally, with the help of client registration in the known settings, areas, or verifiably, with the help of permitting cell phone services to record visited areas. Such data is mutual with different clients who have social connections with each other. A similar data could be abused by LBSN administrator to suggest new POI for the clients. Prescribing unvisited areas poses a significant challenge. It permits to promote organizations with physical nearness efficiently and make income for the LBSN administrator. As an undeniably bigger number of clients participate in LBSNs, the suggestion issue in this setting has pulled in significant consideration in look into and in down to earth applications [20]. The itemized data about past client conduct which is followed by LBSN extricates the issue notably from the other conventional settings. The geographical nature of the locale visited by the previous client conducts as well as the data about client social association with different clients; give a more extravagant foundation to construct an increasingly precise and expressive proposal model. In contrast to conventional methodologies, the calculations don’t exclusively depend on past client inclinations, however, they additionally abuse the social connections of the system as well as the land area of settings. The trial assessment displays that this methodology outflanks customary techniques and the related cutting-edge calculations for suggestions in LBSNs.

4.4 Extraction of User Check-in Behavior with Random Walk for Urban POI Recommender Systems

In order to improvise the nature of shrewd urban life, it’s advantageous that the LBSN prescribes the POIs in which the client might be intrigued. In this manner, efficient as well as viable urban POI proposal system is alluring. In order to model a relevance between the user and there is a Locale of their interest, a User-POI is constructed. In this matrix, the columns denote the POIs and the rows denote the users. Every entry in this matrix indicates a relevance score which is the probability that a particular POI will be visited by a particular user. A Methodology alluded to as Urban POI Walk also termed as UPOI-Walk has been proposed [21]. It takes the social-activated goals (SI), inclination activated expectations (PreI), and notoriety activated aims (PopI) into consideration so as to gauge the likelihood of a client registering to the POI. Its center thought includes building the HITS put together irregular stroll with respect to the standardized registration organize, subsequently supporting the expectation of POI properties identified with every client’s inclinations. To accomplish this objective, a few client POI charts are defined to catch the key properties of the registration conduct persuaded by client aims. In this approach, another sort of arbitrary walk model called Dynamic HITS-based Random Walk has been proposed. It completely thinks about the importance among the clients and POIs from various viewpoints. On the basis of comparability, an online proposal has been made with regards to a POI that the client expects to go to. As far as we could possibly know, this system appears to be the first chip away at POI suggestions for urban that takes client registration conduct propelled by SI, PreI, and PopI into consideration in an area based informal organization information. This likewise handled the issue of extracting client registration conduct in an urban processing condition that is pivotal essential for successful suggestion of POIs in urban zones. The center undertaking of suggestion of POI in the urban regions poses to be advantageously changed to the issue of importance score expectation. The pertinence score of every client-locale pair has been assessed via preparing an arbitrary walk model.

5 POI Based Recommendation Systems Based on Integration of Various Features

There are various techniques and methodologies used to suggest the locale based on the POIs for LSBNs based on the combination of two or more features. Some of them are explained below.

5.1 Graph-Based Approach with Spatial and Temporal Impacts for POI Recommender Systems

This method highlights the problem of time-aware POI recommender system. The recommendation results are time-aware considering the fact that a user often tends to go to different places at different period of time. It is evident that: (a) clients will in general visit close by spots, and (b) clients will in general visit better places in different time period, and during a similar time zone, people will in general occasionally visit similar spots. They have proposed a geographical-temporal influences aware graph, known as GTAG [22]. This graph helps in modeling check in records, spatial impact, and temporal impact. For efficient as well as effective suggestions using GTAG, a Breadth-first Preference Propagation algorithm (BPP) has been developed. This algorithm follows breath first search strategy and returns suggestion outcomes within six (at most) propagation stages. The accessibility of recorded registration information in LBSNs empowers POI proposal administration. This paper has considered the issue of POI proposals that are time-mindful. It takes the transient impact in client exercises into consideration. The GTAG has been proposed to display the registration practices of clients as well as to display a chart based inclination engendering calculation for suggesting POI. These arrangements exploit the topographical as well as fleeting impact along a coordinated way. This exploratory outcome shows that the proposed methodology beats cutting edge POI suggestion strategies considerably.

5.2 Adaptive Approach for POI Recommender System Based on Temporal Features and Check-in Features

Although POI recommendation in LSBNs can help overcome the issue of information overload as it provides personalized location recommendation applications, these systems did not consider the impact of distinct check-in features. This leads to a poor recommendation. This was overcome by the implementation of an adaptive recommendation system called the CTF-ARA [23]. This algorithm fused check-in as well as the temporal features along with the collaborative filtering mechanism based on the user. First, the probability-based statistical analysis methodology was exploited to extract the client activity and similarity information which is the check-in behavior of a particular person. The consecutiveness, as well as the variability factor of the time feature, was also mined. Since the user can be of two kinds either socially active or socially inactive hence clustering was used in order to group users into active and inactive groups. A cosine-similarity of the timeslots smoothing method was utilized to implement POI recommendation, in order to have a method that can work adaptively based on the client’s activity. This approach has experimented with the Gowalla and Foursquare datasets. This method showed better precision and recall than the other recommendation systems seen so far. A novel method has also been implemented by using the geographical, user similarity as well as user activeness influence [8] using Foursquare dataset [4]. It focuses on integration of User check in activity and similarity feature along with the distance feature and time popularity feature.

5.3 Experimental Examination of POI in LSBNs

With the accessibility of the immense measure of clients’ meeting records, the issue of suggestion of POI has been widely considered. It is evident from the research that 60–80 percentile of clients’ tend to make a visit to POIs that were unvisited in the past 30–40 days. POI proposals can incredibly push clients to discover new locales of their inclinations. Various POI prescribed frameworks have been proposed, however, there is as yet an absence of systematical examination thereof. In this method, an overall assessment of 12 cutting edge POI proposal models has been given. From the assessment, a few significant findings have been discovered, in view of which we can all the more likely comprehend and use POI proposal models in different situations. They envision this work to furnish per users with an overall image of the bleeding edge look into on the POI proposal. They speak to the best in class strategies. They spread (I) four well-known suggestion procedures and (II) five kinds of setting data, for example, land influence. And likewise assess the diverse proposal strategies for client inclination displaying in POI suggestions, such as Matrix Factorization, as well as demonstrating techniques for setting data, for example, topographical setting. This assessment will offer bits of knowledge of which technique performs better for every part, for structuring increasingly exact POI suggestion strategies later on. This approach contributes to the first all-around assessment for twelve agent POI suggestion methods.

6 Proposed Work

In this chapter, we discuss the various methods and techniques used in POI based LSBNs based recommendation system. We also propose the implementation of the adaptive POI based LBSNs based recommendation system using user check in influence and the location spatial information. This model can adaptively operate corresponding to the activity of the user. In this method firstly three factors namely time-based POI popularity, 3-dimensional user activity, and the distance feature, are mined [8]. These factors are extracted using a probabilistic statistical analysis technique. Then the user social similarity is extracted. This data are extracted from the historical user datasets. The users are categorized based on their activity as an active and inactive user with the help of a fuzzy c-means clustering algorithm. Lastly, an adaptive recommendation technique is implemented. This method includes the implementation of a 2-dimensional Gaussian kernel density estimation algorithm for the active users and a 1-dimensional power-law function for the inactive users. These are incorporated with the POI popularity and user social influence. The framework of the proposed recommender system is illustrated in Fig. 2.

Fig. 2
figure 2

Proposed framework for POI recommendation

Full size image

6.1 Preprocessing of Data

The users were divided into active and inactive users based on their check in records using FCM clustering algorithm. For each user a three-dimensional activity feature was extracted which include (1) the frequency of user check in, (2) the location check in frequency, and (3) the time distribution of check for each location. After the extraction of these features the social similarity of user was calculated using cosine similarity function. For spatial features, we extracted the popularity of venues with respect to time and the distance of venue from the user location, keeping in mind that most of the users would tend to visit the nearby places which are popular at the time when they want to go out.

6.2 Experimental Results

We have implemented an adaptive location recommender system by using two different strategies based on the user social and check in activity. For socially inactive users we have exploited the one-dimensional power law function. For socially inactive users, we have implemented the two-dimensional Gaussian density estimation functions. We have integrated these two strategies with the popularity of a locale at each interval of time and also with the user social activity similarity among other users.

6.2.1 Dataset

For experimenting the proposed technology, we have used the Foursquare dataset. Foursquare is a large scale LBSN dataset. This dataset is a collection of 227,428 check-in corpus. The description of the data is shown in Table 1. The dataset is very sparse hence the precision and recall results in lower values. The users in the foursquare dataset are divide into active and inactive users using FCM based clustering technique. These users are then used to experiment with our proposed work. The results are compared with the other existing basic recommendation algorithms.

Table 1 Statistics of foursquare datasets
Full size table

6.2.2 Metrics Used for Evaluating the Performance

Precision, recall, and F-measure are used as the evaluation metrics for examining the performance of the proposed technology. Precision is described as the number of recommended locales corresponding to the number of locales present in the test data and is formulated as the average of the precisions for all the timeslots:

$$Precision = \frac{1}{24}\mathop \sum \limits_{t \in T} precision\left( t \right)$$

The recall is the frequency of visitation of venues in the test data in the recommended venues. It is formulated as:

$$Recall = \frac{1}{24}\mathop \sum \limits_{t \in T} recall\left( t \right)$$

The harmonic mean of recall and precision is formulated as the F-measure. In the proposed model we have used \(F_{\beta }\) (\(\beta\) = 0.5). Having the \(\beta\) < 1 puts more emphasis on the precision as well as recall.

$$F_{\beta } = \left( {1 + \beta^{2} } \right) \cdot \frac{precision \times recall}{{\beta^{2} \cdot precision + recall}}.$$

6.2.3 Comparison and Performance Evaluation

This method was implemented and experimented on Foursquare dataset. We evaluated the results by using precision, recall and f1-score metrics. Table 2 shows the performance of our method for the active users. Table 3 shows the performance of our method for the users which are inactive.

Table 2 Performance for active users
Full size table
Table 3 Performance for inactive users
Full size table

Based on the results we can see that as the Number of locations recommended increases the precision decreases and the recall increases. This is because there are too many locations as compared to the number of users. Also by the description of Recall we can see that the denominator is increased as Top-N value but the numerator, that is the accurately recommended locations, are kept constant. And hence the recall increases as the N value in Top-N increases. These results are better than the other state-of-the art recommendation technologies. This result provides us the insight that the proposed technology proves to be more personalized as well as more accurate. The proposed model provides with Top-N for the people in the smart cities to visit according to their preferences, similarity between the other users considering the distance and time-based popularity of the venue. This smart location recommendation technology will provide a secured recommendation without leaking the information to the third party. Though it accesses other similar user database for similarity calculation but at the same time the integrity and privacy will be maintained.

7 Conclusion and Future Scope

Recent research on recommender system show that using social network data can provide improved personalized recommendations with better accuracy of prediction. Due to the lack of knowledge of the cultural sites, users tend to visit only traditional monuments and many charming cultural objects are hidden from them. The emergence of social network has led the individuals in the online community to express their views and share their experiences on visiting a new place and hence this has led to efficient recommendation of new places to the individuals according to their preferences, which thus provides an advance location recommendation system in smart cities which is more accurate. In this paper, we have discussed various techniques and methods used in a POI recommender system for LBSNs. Various factors and features such as social influence, geographical influence, temporal features, etc. that play a major role in the POI recommendation have been discussed. This method will provide people living in the smart cities with an absolute recommendation based on their preferences. This paper also proposes an implementation of already an adaptive POI recommendation algorithm and integration of social influence factor along with this approach. This model deals with the clustering of the users, on the basis of their activities, into active and inactive users using the basic fuzzy c-means clustering algorithm. This makes the recommendation adaptive according to user preference and their activities. In this model, we have implemented POI recommendation using 2-dimensional Gaussian kernel density estimation algorithms (for active users) and a 1-dimensional power-law function (for inactive users). These are integrated with the POI popularity also. This has been experimented with the Foursquare check-in dataset.

In our future scope, we will try to enhance the smart technologies further to provide further accurate solutions for location recommendations in smart cities and help them save money, reduce carbon emissions and manage traffic flows by providing more specific location recommendation. But the complexity of the agenda is hindering its progress but with the collaboration of other technologies like LBSN and POI will bring the smart cities more smart. Hence, we will try to enhance and extend the research on the study of POI recommendation systems. We will further interrogate other check-in aspects, such as category information, social relations, temporal features, etc., and try to integrate these features to provide a satisfactory recommendation. In our future work we would try to use various smart technologies for a more personalized location recommendation for the online smart communities. We will try to explore various applications of these recommendation systems in practice.