29th Sep 2022
16 minute read
Online: AI for subsidized finance
IAFSME – Intelligent ALgorithm for Finance of SME. Intervention co-financed by the E.U. to be worth
on the P.O. ERDF 2014-2020 – Actions 1.5 and 3.8 – TECNONIDI FUND
Access the consultancy via the IAFSME – Intelligent ALgorithm for Finance of SME platform here Access the consultancy via the IAFSME – Intelligent ALgorithm for Finance of SME platform here
.
The term subsidized finance refers to the set of financial instruments used by the legislator at a Community, national, regional or local level to promote the competitiveness and development of businesses. This helps companies to find financial resources on more advantageous terms. In fact, a company can access more tenders and, by planning in time, can obtain more concessions for each type of investment. The requirements for access to the concessions may vary according to the type, the purposes of the tender, the rules set by the issuing body. Natural language processing for subsidized finance helps companies with the automatic suggestion of subsidized finance instruments such as to improve the reputational performance of natural or legal persons who are potentially beneficiaries of the same instruments.
Reputational performance is defined as the quantitative measure of the total expectations, perceptions and opinions developed over time by customers, employees, suppliers, investors and the general public in relation to the qualities of the natural or legal person of interest, verifiable from contents present in social media and mass average. The idea behind the present invention was to adopt a combined set of strategies, while using data analysis techniques and algorithms, which can provide a general model for the automatic suggestion of subsidized finance instruments with performance improvement reputational.
This project uses techniques such as: Named entity recognition (NER); Text categorization; Part of speech tagging.
The NER (Named Entity Recognition), is an NLP task that is placed within a field of study called information extraction. It deals with identifying and classifying the Named Entities present in a text into predefined categories such as people, places, objects, numbers, temporal expressions, etc.
Text categorization or text classification, abbreviated as TC, is always an NLP task that deals with classifying digital texts expressed in a natural language by automatically assigning portions of documents to one or more classes belonging to a predefined “set of classes”. Supervised machine learning approaches can be used to train the model, where it is necessary to train the system through self-learning for instances from which to generate a general model for automatic classification.
Il corpus linguistics part-of-speech tagging (POS tagging o PoS tagging or POST), also called grammatical tag is an NLP task which consists in recognizing a part of text within a period by associating a logical meaning analogously to the logical analysis of the Italian language.
We have obtained a system that searches for tenders from the websites of interest thanks to a crawler and that manages to find the data of interest such as, for example, the date of submission of the application, the deadline for the tender, the admission requirements and the basis necessary documentation. The system in question carries out the reception and identification of the documents present and makes sure that the document is not already present in the database.
It then proceeds to pre-process the data and extract the metadata. The natural language processing for subsidized finance allows an effective identification of security anomalies in the use of data; specifically anomalies present in front-office and back-office activities. It also advantageously combines a classification of inference attempts by the operator, carried out via a convolutional neural network, in such a way as to protect the system from this type of attack, which is difficult to detect using other state-of-the-art methodologies.
The functions of this platform can therefore be summarized as follows: – Provide the end user with a complete set of all financial and reputational strategies suited to the specific characteristics of the client company; – Provide a complete financial diagnostic of the client company’s activity, allowing to trace the causes that determine any financial statement vulnerabilities or insolvency risks; – Allow the algorithms to intelligently adapt to the specific business needs of the end user; – Interface with a higher level internal information system from which to receive information regarding the success rate of financial strategies suggested in similar cases in the past
Sourcing module from databases (crawler) which allows daily updating of the database of subsidized finance opportunities, with retrieval of data from all the sites indicated as sources. There may be cases in which the synergy between the two tasks NER and Text categorization was not sufficient for the extraction of some data, such as for example the identification of the ATECO codes eligible for the tender; to this end, a further NLP task was implemented within the software, i.e. Part of speech tagging. Once the collection of NLP targets has been obtained, it may be necessary to standardize the notations and select the most feasible among various options in order to obtain a coherent database.
The document thus pre-processed is compared with those already present in the database, in order to avoid the filing of duplicates. The project was implemented through a specific online software platform which must perform the complete management of the relationship with the user. In addition, there is a virtual assistance chat interface for the visitor or user: site visitors can contact the customer service or support team in minutes, provide relevant information and start a conversation in real time. This system often saves endless email back-and-forth, increases customer satisfaction, and simplifies the work of sales and support teams.
There are difficulties in accessing tenders and these derive from problems of heterogeneity of the information subjected to analysis, required according to the complete assessment of the company structure and market potential; this forces a laborious phase of manual data collection which at present does not allow the adoption of lean and standardized processes. This results in the impossibility of offering a service at an advantageous price for the customer; in fact, the manual analysis process requires a high share of specialized human resources.
This is a major inconvenience in the case of concomitant orders or where there is not ample time available; the addition of other services to the process, such as for example the evaluation of medium-term financial strategies accessory to the tax credit to obtain a further competitive advantage, entails a further increase in the human resources required.
The natural language processing for subsidized finance consists of a method and system that suggests to the Corporate Management of SMEs financial and business strategies to apply to their business, with the result of a direct improvement in reputational performance. The system and method, based on the company data received from the manager-user and automatically collected from databases, with the use of Business Intelligence, elaborates an innovative company strategy and suggests the tax and financial benefits applicable to SMEs, as well as improvement strategies of the corporate reputation.
This is the first application of our research on natural language processing for concessionary finance. Our results show that these techniques are effective in significantly improving the reputational performance of individuals or legal entities potentially benefiting from the same tools. However, this is only the beginning: we will continue to push these techniques to make it easier for our customers to access more tenders and to be able to obtain more concessions in the future.