Saved in:
| Main Authors: | , |
|---|---|
| Format: | Recurso digital |
| Sprog: | |
| Udgivet: |
Zenodo
2026
|
| Online adgang: | https://doi.org/10.5281/zenodo.20341999 |
| Tags: |
Tilføj Tag
Ingen Tags, Vær først til at tagge denne postø!
|
Indholdsfortegnelse:
- <p><span class="fontstyle0">This research explores the problem of<br>maintaining consistent monitoring of expenditures that is<br>complicated by the necessity for time-consuming data entry<br>and difficulties in making sense of collected financial<br>information. Despite efforts to address this problem, modern<br>financial tracking applications are often limited in their use of<br>OCR technologies and simple charting techniques. These<br>methods may not cope with sophisticated layouts used in<br>receipts and fail to provide any meaningful financial analytics.<br>In response to this need, Expenso – an intelligent native<br>Android mobile application for automated financial operations<br>– is introduced in the current work. Developed using efficient<br>Model-ViewViewModel (MVVM) architecture and utilizing a<br>locally stored Room SQLite database, Expenso is capable of<br>overcoming problems of previous systems by employing a<br>sophisticated Hugging Face AI pipeline for receipt parsing and<br>classification, including TrOCR, LayoutLM, and BART<br>components. This allows for accurate processing of receipt<br>data regardless of its complexity. Furthermore, implementation<br>of Google Gemini AI into Expenso makes it possible for the<br>user to obtain natural language recommendations based on the<br>context of local budgeting. As demonstrated by functionality<br>tests, the OCR accuracy rate reaches 96.4%, whereas the<br>system is capable of operating offline, thus ensuring execution<br>of all functions in the absence of Internet connection.</span> </p>