Who is Selling to Whom – Feature Evaluation for Multi-block Classification in Invoice Information Extraction
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216224%3A14330%2F21%3A00123275" target="_blank" >RIV/00216224:14330/21:00123275 - isvavai.cz</a>
Výsledek na webu
<a href="http://dx.doi.org/10.1007/978-3-030-87802-3_23" target="_blank" >http://dx.doi.org/10.1007/978-3-030-87802-3_23</a>
DOI - Digital Object Identifier
<a href="http://dx.doi.org/10.1007/978-3-030-87802-3_23" target="_blank" >10.1007/978-3-030-87802-3_23</a>
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
Who is Selling to Whom – Feature Evaluation for Multi-block Classification in Invoice Information Extraction
Popis výsledku v původním jazyce
The invoice information extraction task aims at unifying the automatized processing of invoices in structured forms and in the form of a scanned image. Recognizing the pieces of information where a specific value is identified with a keyword (such as the invoice date) is a relatively well-managed task. On the other hand, identification of multi-block information on the invoice, such as distinguishing the seller, buyer, and the delivery address, is much more challenging due to versatile invoice layouts. In this work, we present a new technique of feature extraction and classification to recognize the seller, buyer, and delivery address text blocks in scanned invoices based on a combination of complex layout and annotated text features. The method does not only consider the block positional features but also the relation between blocks and block contents at a higher level. The technique is implemented as a module of the OCRMiner system. We offer its detailed evaluation and error analysis with a dataset of more than five hundred Czech invoices reaching the overall macro average F1-score of 94%.
Název v anglickém jazyce
Who is Selling to Whom – Feature Evaluation for Multi-block Classification in Invoice Information Extraction
Popis výsledku anglicky
The invoice information extraction task aims at unifying the automatized processing of invoices in structured forms and in the form of a scanned image. Recognizing the pieces of information where a specific value is identified with a keyword (such as the invoice date) is a relatively well-managed task. On the other hand, identification of multi-block information on the invoice, such as distinguishing the seller, buyer, and the delivery address, is much more challenging due to versatile invoice layouts. In this work, we present a new technique of feature extraction and classification to recognize the seller, buyer, and delivery address text blocks in scanned invoices based on a combination of complex layout and annotated text features. The method does not only consider the block positional features but also the relation between blocks and block contents at a higher level. The technique is implemented as a module of the OCRMiner system. We offer its detailed evaluation and error analysis with a dataset of more than five hundred Czech invoices reaching the overall macro average F1-score of 94%.
Klasifikace
Druh
D - Stať ve sborníku
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
<a href="/cs/project/LM2018101" target="_blank" >LM2018101: Digitální výzkumná infrastruktura pro jazykové technologie, umění a humanitní vědy</a><br>
Návaznosti
P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)
Ostatní
Rok uplatnění
2021
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Údaje specifické pro druh výsledku
Název statě ve sborníku
SPECOM 2021: 23rd International Conference on Speech and Computer
ISBN
9783030878016
ISSN
0302-9743
e-ISSN
1611-3349
Počet stran výsledku
12
Strana od-do
250-261
Název nakladatele
Springer
Místo vydání
St. Petersburg, Russia
Místo konání akce
St. Petersburg, Russia
Datum konání akce
1. 1. 2021
Typ akce podle státní příslušnosti
CST - Celostátní akce
Kód UT WoS článku
—