Semantic Enterprise Search

Enterprises need dedicated technology that overcomes the limitations of search engines.

Modern systems in Information Technology are used to deal with a huge amount of information as the data stored and daily produced in the Web and Social Networks. Most human knowledge is represented and expressed using natural language and the proper application of Natural Language Processing (NLP) methods is crucial in exploiting such data. Natural language interpretation is used in Reveal for recognizing the core information contained in documents and to store it properly in order to support fast, accurate and proactive document retrieval processes.

Organizations depend on such retrieval mechanisms in several operations:

• Locating useful technical information within its archives or globally on the Web.
• Locating organizational and logistics information within its own information ecosystem, usually distributed across different and often independent systems and servers.
• Gathering all information about a process (e.g. auditing an internal process or designing a new product)
• Summarizing it in suitable and exhaustive summaries.

All these tasks are highly organisation dependent and are also specialized towards specific sectors of an organisation so that independent areas of the company require different search strategies. The above overall requirements form the basis of a dedicated technology needed by enterprises that overcome limitations of search engines, called Semantic Enterprise Search (SES)

In support of the essential SES processes, more complex structures expressing meaningful information are made available

Traditional Information Retrieval (IR) systems deal with representation, storage, organization of, and access to information items, e.g. documents, through the coding of document contents just exploiting word occurrences (i.e. at the so-called full-text level). These methods are fully lexicalized and well established in traditional search engines. However, the advances in this technology come from richer document representations able to express (store and index) the text contents at the grammatical and semantic level: in this way, more complex structures expressing meaningful information are made available, in support of the above essential SES processes (from search to conceptual document gathering and summarization) so crucial for modern enterprises.

Revealer, the integrative SES solution

Reveal offers SES solutions that integrate domain specific NLP, machine learning solutions for Information Extraction and document classification as well as search for a comprehensive SES framework, called Revealer.

Reveal supports linguistic processing over different types of texts (from norms to Web documents, from short questions to Social Media posts and micro-posts, e.g. tweets) in order to offer an integrated IE environment over distributed and independent text archives. This allows the support to the indexing and fast retrieval of a variety of conceptual phenomena, such as entities, persons, events and relational information, crucial in SES functionalities. Finally, the retrieval functions are combined with browsing capabilities that see the document archive as a large document graph and support conceptual search and navigation: the first, to locate the useful information and second to use it as the suitable take-off for navigation. This realizes an entire information ecosystem integrating texts and concepts altogether. Examples of navigation within such an information ecosystem are reported in the figure below, within a financial domain: a large international bank.

Organisational units of the bank (blue nodes) are connected to the documents (orange nodes) they are responsible for. Also processes (green nodes) are linked to the former according to relations with documents (e.g. norms), i.e. the constraints these impose on individual processes.

Another example is the dendrogram-like chart hierarchically representing independent topics of discussion that are dynamically derived from daily tweet streams: in the example, discussion about the Giulio Regeni case (June 2020) are clustered separately from other school-oriented topics (such as June 2020 Maturity exams). This structure is evoked entirely by the system and allows novel forms of querying and browsing within the very large text collection corresponding to tweet streams.

whereas the (bottom) layers related to Language Processing and Indexing are used to support Data Mining over the document base as well as advanced querying and navigational capabilities. The front-ends thus inherit all the semantics from the lower layer at the individual document level (through NLP over texts) to the overall (summarized) information and knowledge as emerging from the entire document collection (already processed thus interpreted) via Data Mining abilities to foster knowledge discovery and analytics.

Strong integration with Dialog Agents

The Reveal SES environment also supports natural language querying modalities for speech-based or text-driven Question Answering over the indexed contents. This supports a strong integration with dialogue (conversational) agents able to act as friendly interfaces for complex searches (e.g. about compliance problems, or customer relationship operations). The main objective of a Question-Answering system is to automatically answer a question posed in natural language or retrieving that answer in a document collection: the key problem is when fine-grained phenomena are targeted the lexical information alone is not sufficient and the full exploitation of Reveal NLP technologies comes into place.