Approximate Nearest Neighbour Search on Privacy-aware Encoding of User Locations to Identify Susceptible Infections in Simulated Epidemics
| Autor: | Biswas, Chandan, Ganguly, Debasis, Bhattacharya, Ujjwal |
|---|---|
| Rok vydání: | 2020 |
| Předmět: | |
| Druh dokumentu: | Working Paper |
| Popis: | Amidst an increasing number of infected cases during the Covid-19 pandemic, it is essential to trace, as early as possible, the susceptible people who might have been infected by the disease due to their close proximity with people who were tested positive for the virus. This early contact tracing is likely to limit the rate of spread of the infection within a locality. In this paper, we investigate how effectively and efficiently can such a list of susceptible people be found given a list of infected persons and their locations. To address this problem from an information retrieval (search) perspective, we represent the location of each person at each time instant as a point in a vector space. By using the locations of the given list of infected persons as queries, we investigate the feasibility of applying approximate nearest neighbour (ANN) based indexing and retrieval approaches to obtain a list of top-k suspected users in real-time. Since leveraging information from true user location data can lead to security and privacy concerns, we also investigate what effects does distance-preserving encoding methods have on the effectiveness of the ANN methods. Experiments conducted on real and synthetic datasets demonstrate that the top-k retrieved lists of susceptible users retrieved with existing ANN approaches (KD-tree and HNSW) yield satisfactory precision and recall values, thus indicating that ANN approaches can potentially be applied in practice to facilitate real-time contact tracing even under the presence of imposed privacy constraints. Comment: 8 pages, 8 figures |
| Databáze: | arXiv |
| Externí odkaz: |
|