Travel expenses represent up to 7% of organisations overall budget. Existing expenses systems are designed for reporting expenses types and amount, but not for understanding how to save and spend. We present a system, manipulating semantic web technologies, which aims at identifying, explaining, predicting abnormal expense claims by employees of large organisations in 500+ cities. Abnormality Detection Description: This service retrieves abnormal accommodation price in a given fcityg, fcountryg for year 2015 worldwide.
Time and expense in most companies is the process of recording and tracking hours worked and expenses as they relate to projects. $546 Billion worldwide has been estimated to be lost because of non spend optimisation in the process of managing expenses at organisation level. We address the problem of spend optimisation i.e., determining how to save and spend. Our system, AIFS (Accenture Intelligent Finance System), aims at (i) identifying, (ii) explaining and (iii) predicting abnormal expenses e.g., flight, accommodation, entertainment (denoted as abnormalities) in 500+ cities. The system is mainly used by (i) travel and expenses business owner to better manage spend optimisation, (ii) expenses auditor for tracking abnormal expenses, and (iii) internal travel system administrator for defining expenses policy based on the reasoning results.
Our system is exposing a semantic description of causation between external events and expenses type. It leverages open data (e.g., events), LOD and reasoning functionalities to expose causation using semantic representation in RDF. DBpedia and Wikidata are used by our system for representing heterogeneous data together with explanation, and then by our back-end reasoning engine for semantic association. Our system exposes three core functionalities: abnormality (i) detection, (ii) explanation, and (iii) prediction. They cover the particular case of accommodation price variance over time in 500+ cities. All functionalities consume JSON inputs and expose back JSON, RDF and JSON-LD with contexts mapped to the LOD cloud. JSON-LD has been considered for further easy integration in any application. 2.1
Output: Abnormal accommodation prices are detected in 500+ cities. The JSON description captures the type of expenses, its date, its severity level (scaled from 1 to 5 as the most severe), its average and observed amount in US dollar. The RDF description exposes semantics of the search and abnormalities. For instance, cities and countries are contextualised using their DBpedia resources.
Back-end Technology: A standard statistical analysis of data has been conducted to detect abnormal expenses. An analysis of 300,000+ unique travellers in 500+ cities worldwide with a minimum and maximal number of respectively 2,521 and 24,800 expenses per city recorded for 2015 has been initially performed. 2.2
Description: We retrieve semantic explanation of abnormal price of accommodation in a given fcityg and fcountryg between fdate startg and fdate endg. 1 2 3
Output: Semantic explanation, contextualised using DBpedia and Wikidata, of abnormal accommodation price is provided in 500+ cities. Contributing factors (scaled from 0 to 100), categorised as frauds, external events, city capacity, season together with their types (e.g., music category for social event) are provided c.f. Figure 1. Besides RDF schema, an internal ontology, characterising links across categories, is also used.
Back-end Technology: Abnormality explanation has been achieved by applying core
principles of explanatory reasoning [
2.3
Description: The system retrieves prediction of accommodation price and its explanation when the price is abnormally high.
Back-end Technology: Prediction of abnormality has been achieved by applying core
principles of predictive reasoning [
Data and Vocabularies
Background Data in Use: We define background data as data consumed silently by
the system with no user specification. Only relevant background data (i.e., data which
could be mapped to DBpedia, Wikidata for semantic comparison cf. [
Data
Source 300,000+ unique travel- Min. and max. number lers in 500+ cities of respectively 2,521 and recorded for 2015 24,800 expenses per citya
Social events e.g., music event, political event .93 (complete) .41 (aggregated)
Private a https://www.eventbrite.com/api b http://api.eventful.com c http://wiki.dbpedia.org/Datasets d https://www.wikidata.org/wiki/Wikidata:Database download e http://54.194.213.178:8111/ExplanatoryReasoning/ontology/categories.n3 4 Conclusion and Plan for Extensibility We presented AIFS (Accenture Intelligent Finance System) i.e., a system for interpreting, explaining and predicting abnormally high expense using accommodation price as abnormality in 500+ cities. The system, exposing semantic descriptions of causation between events and expenses type, is mainly used by (i) travel and expenses business owner to better manage spend optimisation, (ii) expenses auditor for tracking abnormal expenses, and (iii) internal travel system administrator for defining expenses policy based on the reasoning results. Semantics has been crucial to contextualise and interpret expenses and their amount.