Point of Interest (POI) categories can facilitate a number of services, such as location-based search and place recommendation. However, such information can be incomplete and/or incorrect, especially in crowdsourcing environments. In the literature, automatic category imputation has been suggested to tackle this problem, showing that contextual information is vital for increasing the quality of such predictions. To this end, users' check-in data, and most particularly location and time of visit, is often used as the notion of context. In this work, we propose a method that considers culture as a contextual parameter. Contrary to existing methods, our approach does not require access to user data. We illustrate the feasibility of our method by performing experiments on data from Foursquare, a global location-based social network.
Point of Interest (POI) categories can facilitate a number of
services, such as location-based search and place recommendation.
However, as discussed in recent work, categories, especially in
crowdsourcing environments, can be incomplete and/or
incorrect. Automatic category prediction has therefore been proposed
to remedy this problem and impute missing categories [
Recent advances in POI categorisation have shown that
contextual information is vital for increasing the quality of automatic
POI category prediction [
• To our knowledge this is the first study of a multi-lingual, global location-based social network database related to 1.1
Our company Naver, provides, among other things, locationbased services. Good quality POI data is thus of major importance. In this context, in Naver Labs Europe, we have been exploring automatic multi-lingual methods for completing and correcting POI semantic tags found in Foursquare’s database, a global crowdsourced location-based social network1. The scope of our work is to eventually support: • A user that is not familiar with local culture to discover appropriate POIs in the vicinity of her/his position. If POIs are not categorised appropriately, the user can not easily search for them and they will not be included in the search results. In addition, proper POI categorisation could also help in recommendation. Most of the work on Point-of-Interest category prediction has taken place in the context of Location-Based Social Networks.
Ye et al. [
Krumm [
The type of user data explored in the state-of-the-art include the number, frequency, time and duration of user check-ins, and sometimes demographic information (e.g. age range, gender). This data is usually combined with the location of the POI and its proximity to other POIs. 1We got access to this data thanks to an agreement between Naver Labs and Foursquare.
He et al. [
Jiang et al. [
A number of works have been carried out on location
prediction based in social streams, e.g. Twitter, where the main research
interest is using noisy and short text for classification. For
instance, Cano et al. [
To summarise, as mentioned in the Introduction, all the aforementioned approaches assume access to user data at training time and have not dealt with the notion of user’s culture. 2.2
Recent work had highlighted the importance of modelling users’
culture in recommendation and information search [
To our knowledge, in the area of automatic recommender
systems, Zangerle et al. [
Our goal is to predict POIs’ categories that are appropriate to a specific culture. For instance, a typical place found in the database of Foursquare is "La Table du Ramen"3, which is located in Paris, France. The category found in Foursquare’s database is Japanese Restaurant, which may be suficient for the local culture. However, a Japanese would expect the POI to be categorised at a much more fine-grained level, as e.g. Ramen Restaurant or as a Noodle House. The objective is to automatically predict such categories, according to one’s culture.
We consider that POI category prediction in our context is
equivalent to the problem of completing a specific attribute of
the dataset, the one that represents categories of POIs, based
on data from the remaining attributes. Formally, a POI p should
have an attribute that includes an ideal, complete and correct,
category labelset i.e. set of relevant labels L ⊂ Λ, where Λ =
l1, ..., lm is the set of all possible labels, while the rest of the
attributes are represented by the set A. However, in practice,
the set of labels attributed to p, what we call thereof the
observed labelset Lo ⊂ Λ, can be incomplete and/or incorrect.
2Hofstede et al. [
In this work, we assume that there are several "culture spen cific" labelsets, such that S Lj ⊆ L, where n is the maximum j=1 number of cultures that are represented by the labels in Λ and n < ∞. For instance revisiting our example, if we assume that L = {Japanese Restaurant , N oodle House, Ramen Restaurant }, and we have at least two cultures French and Japanese with corresponding labelsets L1 and L2, then it could be that L1 = {Japanese Restaurant , N oodle House } and L2 = {N oodle House, Ramen Restaurant }, while Lo = {Japanese Restaurant }.
We denote by y = (y1, ..., ym ) an m-dimensional binary vector
where yi ∈ [
We assume that C ∪ N = A where C stands for the set of culture-related attributes, and N the set of culture-independent ones. Considering again our example, the C could be instantiated by the country where the restaurant is located, and N by the name "La Table du Ramen".
We denote by x, xC , xN the vectors that represent correspondingly A, C, and N , such that the observed p in the dataset, is defined as
p = {x, yo } = {xC , xN , yo } Our objective is to make culture-specific predictions such that for culture c
p = {x, yc } We thus formulate our goal as a multi-label classification problem where we want to find a classifier bc : X → Y where X is the input space (all possible attribute vectors) and Y the output space (all possible labelset vectors), such that yc = bc (x ). 4
The category of POIs, especially in a location-based social network, is related to the cultural profile of the users that visit it. This has been proven via the inclusion of user profiles in related work (c.f. Section 2). However, instead of accessing user information to discover such profiles, we use the observation that the majority of POIs are categorised in a manner that reflects local culture in location-based social network databases. For instance, as shown in Fig. 1, we find in Foursquare that restaurants selling noodle dishes4 are usually categorised as Asian Restaurant in France, while Ramen Restaurant is by a large margin the most popular category in Japan. 4The token "noodle" is explicitly mentioned in the POI name. then at inference time we can replace the corresponding inputs according to the target cultural profile. For instance, if at training time we use the country in which the POI is located as an input parameter to our model, at inference time we can replace the value of this parameter with the target country, simulating what would happen if the same POI was located in the target country instead. We can thus generate culturally-appropriate predictions and complete the database ofline. Revisiting the above example, we can assume that some of the Ramen restaurants located in Japan, would be categorised in a diferent manner, e.g. as Noodle Houses, if the value of the country was changed to France5. An overview of the proposed method is shown in Fig. 2.
Based on the discussion above, we reformulate our problem, and look for a classifier bc such that yc = bc (xc ) where xc is a culture-specific variant of the input.
To find bc , we follow a standard approach and transform our
problem into finding a real-valued vector function f : X → S ∈
[
To learn the binary models we perform the following steps. • Attribute selection: We use the name and spatial geocoordinates of the POIs. • Vectorisation: This step includes transforming the attributes in a form that can be treated by the classifier, as explained in Section 4.1. • Training: A model that learns to predict the probability of li being a correct label given x is computed in this step. As explained in the previous paragraph, our problem is casted as a supervised machine learning problem. Details are provided in Section 4.2. • Inference: Whether li is a correct label for culture-specific variants of x is computed in this step. Details are given in Section 4.3. 4.1
Categorical variables. We represent them with one-hot
encoded embeddings, as usually reported in the literature.
Sequential variables. Biessmann et al. [
Spatial variables. Geographical coordinates are the most
important spatial attributes that characterise a POI. For instance,
latitude and longitude are two of the most frequently used
geographical coordinates. The predominant way of modelling
coordinates is to discretise the input space [
Note: Other cultural variables. This is an optional category, which can include other parameters related to culture. For instance, to determine socio-cultural context, opening hours and the price range of corresponding services may also be important. Both of these variables could be discretised and considered categorical variables. 4.2
Once we vectorise our attributes, as explained in the previous section, we use a concatenation layer to combine them.
If a is a POI attribute such that a ∈ A and ϕa (xa ) ∈ ℜDa is the attribute specific vectorisation function, where Da denotes the dimensionality associated with the attribute a, then the final input vector is a concatenation of all vectorised individual attributes: x˜ = [ϕ1 (a1), ϕ2 (a2), ..., ϕn (an )] where n denotes the number of attributes. We feed this to a dense layer
h = relu[W hx˜ + bh ] After applying a dropout layer, we then calculate
p (y |h, θ ) = siдmoid[W h + b] where θ = (W , b, W h, bh ) are learned parameters of the model. 1 siдmoid (s ) denotes the logistic function f (si ) = 1+esi . The parameters θ are learned by minimising the binary cross-entropy loss function. 4.3
The multi-label model learned is given culture-specific inputs at
this step. It then generates for each label a probability score. To
get from that the corresponding set of accepted labels, a constant
can be applied as threshold (usually this is 0.5) [
Shop & Service Outdoors & Recreation Professional & Other Places Arts & Entertainment Travel & Transport College & University Nightlife Spot Event Residence with 336 categories distributed in 5 levels. The least developed one is Residence having only 4 subcategories, over 2 levels7.
Label distribution. We have extracted a sample dataset from the existing Foursquare database for our experiments. We used the most well developed root category, Food, as seed, and took only POIs with a high reality index. The distribution of the categories is similar to the one found in the original hierarchy. To better understand the dataset we provide the overall distribution of POIs and the one of the top categories in Table 2. The dataset is skewed in terms of the POI instances attributed to each category, with the first 10 top categories having more than 40% of the POIs attributed to them.
Point-of-Interest attributes. . POI attributes include its name,
and the latitude and longitude, transformed into the
representation discussed in Section 4.1. It is important to note that
contrary to completely freely crowdsourced POI databases such as
OpenStreetMaps [
Dataset creation. To generate training, development, and test data, we used approximate stratified sampling. The goal was to maintain the distribution of positive and negative examples of each label by considering each label independently. Consequently, we allocated the POIs proportionally into 40% for training, 10% for development, and 50% for testing purposes8. 8We kept a relatively large percentage of the data for testing pusrpose, in order to have a large enough sample of POIs belonging to long tail categories, where cultural diferences may be more obvious.
Model. We use a neural architecture with one hidden
dense layer, followed by a dropout layer, and the output layer.
We use the Rectified Linear Unit as the activation function of
the hidden layer. The dropout rate is set to 0.3. The loss we use
is binary crossentropy. We have set an early stopping criterion
for the training based on a pre-defined threshold that takes into
account the delta of the loss between two consecutive epochs. For
all sequential features we applied a length of 50. For the LSTM
layer we set the dimensions of the embedding layer vector space
to 128 and the number of the LSTM hidden units to 128. The
LSTM has a recurrent droupout rate of 0.3. Experiments were
run on a single GPU instance (1 GPU with 16GB VRAM, 4 CPUs,
with 256GB RAM). Training was performed with a batch size
of 32. We used the Adam optimiser with the default parameters
recommended in the paper [
• The model learned that some categories are by design (as defined by Foursquare) allowed only in specific countries e.g. Acai House and Churrascaria only in Brazil. • The predictions reflect, to the best of our knowledge, local culture reasonably well e.g. Bistro is popular in France and Souvlaki Shop in Greece. • Semantically similar categories are found by the classiifer between diferent cultures, even when the original category is specific to one culture only. For instance Pastelaria, a typical Brazilian and Portuguese POI category, is predicted in other cultures as Bakery and/or Snack Place, which is correct. Similarly, Churrascaria is classified as
Latent similarities discovered by the model are questionable though in some cases. For instance, Souvlaki Shop, a prominent category in the Greek culture, is correlated to the category Kebab Shop in other cultures. In reality, in a number of aspects there are strong similarities: for instance the form of the corresponding sandwiches (pita-based and circular), the meat that is cooked on a spit, and the "fast food" type of food delivery. However, the former is made traditionally out of pork and the latter does not have pork, which is an important diference. In future work, we plan to tackle this aspect further. 6
We have presented a new method to predict, in a culture-specific manner, POI categories, without requiring access to user information. To achieve that, we simulate user information, by replacing culture-related training inputs in an appropriate manner, at inference time. For instance, the country where a POI is located can be replaced at inference time the by the nationality of the user. We have performed preliminary experiments on data of a global location-based social network, Foursquare, that give us promising results. In future work, these results will be further verified with user studies.
Original data
US
TR
GR Acai House 1488 43 29 0 1534 0 0 Note:Except for Brazil in the rest of the cultures the same POIs are categorised as Snack Place, Juice Bar, Dessert Shop. According to Foursquare’s documentation Acai house is a category only supported in Brazil.
Bistro 879 470 3310 351 1162 95 10 Note: Bistros predicted using the French culture are tagged in the original data as: Café, Bar, Gastropub, Diner. Corresponding predictions in other cultures are: Café, Wine Bar, Bar, Gastropub.
Brasserie 18 0 91 0 3 0 0 Note: In other cultures Café is the main predicted category for the same POIs (or there is no prediction at all). In the silver standard the POIs are also categorised as Bistro or Café.
Café 126665 108908 87381 63223 108837 148050 236436 Note: In the US culture a lot of Cafes seem to be categorised as Cofee Shops instead. In the Greek culture Cofee Shop, Breakfast Place, Dessert Shop, Bar, Tea Room POIs are categorised as Café (which is actually representative of the culture). Cofee Shop 54291 49069 Note: As explained in the previous row.
45769 68102 50513 52742 23629 Churrascaria 557 0 0 0 674 0 0 Note: Churrascaria is a Portuguese/Brazilian BBQ. In other cultures the majority of the same POIs are classified as BBQ Joint and a small percentage as Steakhouse (especially in the US).
Creperie 978 775 2968 932 1138 917 1740 Note: Creperies are obviously common place in the French culture. In the US and KR ones the same POIs are rather categorised as Dessert Shop or Breakfast Shop. In the BR one in addition to Dessert Shop we find also Pastelaria 9. Dessert Shop 15164 19422 4747 9651 15460 21941 13736 Note: In the French culture Dessert Shop POIs are rather classified as Café, Bakery, Creperie, Pastry Shop, Chocolate Shop. In the US one we have to note the large number of POIs categorised as Ice Cream Shop, Frozen Yogurt Shop, Candy Store. Diner 2590 2289 2388 3980 1539 1593 100 Note: Some of the POIs categorised in the US culture as Diner, are mainly categorised in other cultures as Café or Breakfast Spot or there is no prediction (the diference is really big with Greece where almost all of them are categorised as Café). Friterie 656 7 4864 2 419 4 62 Note: The model has learned that Friterie is a culture-specific category (supported in FR, BE, NL in the Foursquare database). It is interesting to note that in the US culture the same POIs are categorised as Burrito Place, Taco Place, Food Truck, in the TR one as Kofte Place and in the GR one as Snack Place. They all seem to share a fast food aspect.
Meyhane 854 0 0 0 0 2157 0 Note: The model has learned that Meyhane is TR culture-specific category. In other cultures Meyane POIs usually do not get any prediction.
Pastelaria 340 1 0 0 898 57 0 Note: The model has learned that Pastelaria is BR (and Portuguese) culture-specific category. In FR culture categorised as Creperie, Snack Place, Bakery, US:Bakery,Snack Place, KR:Bakery, Dessert Shop, Snack Place.
Pastry Shop 60 5 307 0 0 88 0 Note: The model has learned that Pastry Shop is more frequent in FR (i.e. Patisserie). In other cultures we mainly find Dessert Shop or Bakery.
Souvlaki Shop 94 0 0 0 0 0 2434 Note: The model learned that Souvlaki Shop is specific to GR culture. In other cultures and in the silver standard, the same POIs are categorised mainly as Fried Chicken Joint and BBQ joint. There is a strong correlation to Steakhouse as well, or more precisely to Kebab shops that are also classified as Steakhouse in the silver standard.
Sports Bar 67 9 2 100 4 5 0 Note: The model learned that Sports Bar is more frequent in the US culture. In the silver standard, the same POIs are categorised also just as Bar and/or Wing Joint. Furthermore, the model learned a strong correlation between the category Wings Joint and sports Bar - the two categories are predicted together 77% of the time.
Takoyaki 186 130 45 34 22 44 30 Place Note: Takoyaki Place POIs do not get any prediction most of the times in other cultures except for KR. 86% of the predicted POIs are correct according to the silver standard.