WORK PACKAGE 1
This work package will be led by metaphacts and involve all partners. The main goals of this work package are twofold. First, we will specify the exact functional requirements of the platform components based on user requirements. To achieve this goal, the requirements of the two use cases as well as current development on the research scene will play a central role. The consortium will carry out an analysis of the state of the art and define the target state after the completion of the project. The requirements elicitation and furthermore the requirements specification will be performed on the basis of these analyses. In a second step, the consortium will develop a suitable technical architecture to fulfill these requirements.
WORK PACKAGE 2
This work package will be led by the University of Paderborn and involve all partners. The main goals of this work package are twofold. First, we will develop an approach to compute the veracity of single facts based on both textual evidence and corroborating facts found in the same knowledge graph. To achieve this goal, the consortium will develop novel solutions based on supervised explainable machine learning algorithms.
The second goal is to check not only single facts but a set of facts (ranging from small sub-graphs to large knowledge bases) for their veracity, as a simple approach relying on the check of every single fact in the knowledge base is not feasible for large knowledge bases. Hence, we will develop an approach able to detect central the facts of a knowledge base. These will then be checked for veracity and used to approximate the total veracity.
Both goals include the gathering of evidences which either support a fact or refute it. These evidences can be used to generate explanations for humans in work package 3.
WORK PACKAGE 3
his work package will be led by the University of Paderborn and involve all partners. Its main goal is the generation of explanations of the fact checking decisions the FROCKG framework carries out. The explanations have to be understandable by a typical customer of the use case domain.
Therefore, this work package is mainly based on the evidence extracted within work package 2. This evidence will be summarized to represent a shortened but complete picture of all evidence which has been found. Additionally, the trustworthiness of single evidence sources as well as the veracity of knowledge bases have to be included into the explanation.
WORK PACKAGE 4
This work package will be led by metaphacts and involve all partners. Its main goal is to develop the workflows and support the user in her knowledge graph curation and data acquisition tasks. The curation workflows must take into account the suggestions generated by the fact checking algorithm and the explanations provided by the techniques developed in Work Package 2 and Work Package 3 as well as support manual fact checking by human users. In particular, the tasks of this workflow will include supporting and managing human argumentation, developing intelligent editing forms and components incorporating results of fact checking algorithms and providing explanations, and realize editing workflows.
WORK PACKAGE 5
This work package will be led by SIRMA and involve all partners. Its main goal is to use the results of automated fact checking (done with the help of Work Package 2 tools) and manual editing and curation of the knowledge graph (using Work Package 4 workflows and UI) as an input to improve knowledge extraction applications. The targeted improvements are twofold: On the one hand, curated facts (both on the concept and instance level) can be used to update enrichment gazetteers. On the other hand, training corpora can be updated and ML-based components can be retrained to improve their F-measure. Such continuous adaptation of knowledge extraction is currently achieved only through manual curation and evaluation of the results of information extraction over gold standard corpus documents by domain experts. This is a tedious process that also requires experts to rely on technical staff that can perform necessary CRUD operations on the knowledge graph. Leveraging automated fact checking algorithms and the visual tools for curation and management of the knowledge graph will eliminate the need for technical knowledge and allow domain experts to actively participate in the incremental improvement of the knowledge extraction application at a lower cost.
To achieve its goal, this work package will produce generic tools that enable updating and retraining of information extraction components on the basis of automatically validated or user-validated facts. These tools will then be integrated in the Semantic tagging solution of SIRMA to complete the workflow of semantic enrichment.
WORK PACKAGE 6
The aim of this work package is to evaluate the FROCKG framework within an enterprise setting. The FROCKG use cases will be provided by customers of metaphacts, Zazuko and SIRMA AI from the cultural heritage and financial domain. Both domains being data-intensive and relying significantly in making value out of data by channeling the right data to the right expert at the right time. The work package will be led by SIRMA.
WORK PACKAGE 7
This work package will be led by metaphacts and involve all partners. Its function will be the coordination of the whole project and the dissemination of project results.