Indoor Location Based Recommender System
Author
Jaiswal, Spardha
Term
4. term
Education
Publication year
2018
Submitted on
2018-09-25
Pages
52
Abstract
This thesis investigates how to adapt recommendations in a shopping mall to a user’s current indoor position and interests. The goal is to design a location-aware recommender that combines three inputs: static data (the mall floor plan and a store/item catalog with keywords), dynamic data (the user’s indoor location), and the user’s historical web browsing. Indoor positioning is achieved via a WiFi fingerprinting method that partitions the floor plan into regions (vicinities) and stores ranges of RSSI signal strengths for the five access points with the strongest signals in each region, rather than full measurements at many points. At run time, the device’s WiFi scan is matched against this radiomap to identify the user’s vicinity. The recommender extracts keywords from the user’s web history, retrieves keywords for stores and items in the vicinity, and matches them to rank and present a limited list of relevant items along with store names; the list can update as the user moves. The thesis details the system architecture, data model, and the localization and recommendation pipelines, and outlines a prototype to support implementation and testing. The provided pages focus on methodology and design; specific evaluation results are not included in this excerpt.
Dette speciale undersøger, hvordan anbefalinger i et indkøbscenter kan tilpasses en brugers aktuelle indendørs position og interesser. Projektets mål er at udvikle en lokationsbevidst anbefaler, der kombinerer tre typer input: statiske data (centrets plantegning og butiks-/varekatalog med nøgleord), dynamiske data (brugerens indendørs position) og brugerens historiske websøgninger. Indendørs lokalisering sker via en WiFi-fingeraftryksmetode, hvor gulvplanen opdeles i regioner (nærzoner), og der lagres intervaller af RSSI-signalstyrker for de fem adgangspunkter med højest signal i hver region i stedet for fulde måleserier. Ved kørsel matches enhedens aktuelle WiFi-aflæsning mod dette radiokort for at bestemme brugerens nærzone. Anbefaleren udtrækker nøgleord fra brugerens webhistorik, henter nøgleord for butikker og varer i nærzonen og matcher dem for at rangere og præsentere en begrænset liste af relevante varer sammen med butiksnavne; listen kan opdateres, når brugeren bevæger sig. Specialet beskriver systemarkitektur, datamodel, procesforløb for lokalisering og anbefaling samt en prototype som grundlag for implementering og test. De første kapitler præsenterer metode og design; konkrete evalueringsresultater fremgår ikke af det medfølgende uddrag.
[This apstract has been generated with the help of AI directly from the project full text]