Do You Like What I Like? Similarity Estimation in Proximity-Based Mobile Social Networks

Abstract

While existing social networking services tend to connect people who know each other, people show a desire to also connect to yet unknown people in physical proximity. Existing research shows that people tend to connect to similar people. Utilizing technology in order to stimulate human interaction between strangers, we consider the scenario of two strangers meeting. On the example of similarity in musical taste, we develop a solution for the problem of similarity estimation in proximitybased mobile social networks. We show that a single exchange of a probabilistic data structure between two devices can closely estimate the similarity of two users -without the need to contact a third-party server. We introduce metrics for fast and spaceefficient approximation of the Dice coefficient of two multisets -based on the comparison of two Counting Bloom Filters or two Count-Min Sketches. Our analysis shows that utilizing a single hash function minimizes the error when comparing these probabilistic data structures. The size that should be chosen for the data structure depends on the expected average number of unique input elements. Using real user data, we show that a Counting Bloom Filter with a single hash function and a length of 128 is sufficient to accurately estimate the similarity between two multisets representing the musical tastes of two users. Our approach is generalizable for any other similarity estimation of frequencies represented as multisets.

On gossip-based information dissemination in pervasive recommender systems

“…Similarity data gets propagated to nearby peers and therefore has to be privacy-preserving. Privacy-preserving similarity comparison can among others be performed on item vectors [4] as well as texting data [15]. • Context Data: Data that characterizes the encounter such as location, time, weather, or peer activity (running, eating, commuting) that can be sensed (for example via sensors) or retrieved (for example from the web) [6,30].…”

Affinity: A System for Latent User Similarity Comparison on Texting Data

“…After updating the global DF vector, a truncated DF vector corresponding to the words submitted by the user, and the total number of users is returned to the smartphone. The user can now calculate tf−idf features for all of his/her used words according to Equation (1), which are then added to his/her BoW vector. The WMD calculates distances between two (word vectors, word weights) pairs, which are commonly referred to as signature.…”

“…6 Techniques from differential privacy can be used to add jitter to the true word vectors to obfuscate knowledge on the hidden word embedding model. of the vocabulary word v in R. The user DF vector can then be formalized as DF user (A) = [1] v∈V . The user DF vector is sent to the backend, where it updates the global DF vector…”

“…The truncated global DF vector together with the number of total users #U are sent back to the smartphone. We can now calculate the tf−idf values for all vocabulary words in V according to Equation (1). We then pick the top 2 tf−idf valued vocabulary words, for instance country and nation, and send them to the backend.…”

“…Especially with the proliferation of smartphones and the popularity of location-based services, there is also a growing interest in more dynamic scenarios. Combining the ideas of ubiquitous computing and social networking, it is possible to connect to similar and close by people who not necessarily are close in the social graph [1], [2]. In the field of psychology, the principle of homophily states that individuals tend to bond if they are similar.…”