Accern's NoCodeNLP platform allows users to create workflows that extract and generate structured data from textual data. Below we present the terminology used to designate each of the components of the output data.

We designate any individual textual information, such as news articles, tweets, financial documents, blog posts, etc., as a document.

A document cluster is a group of contextually similar documents. A new document cluster is formed when our document clustering algorithm cannot find a related document cluster for a document. A document cluster can contain multiple documents. However, a document can be part of only one document cluster.

We define signals as text parts that provide similar information about an entity/event or both. A document can contain multiple signals.

We define any information extracted directly from a document as its metadata. Metadata includes information like hyperlinks, publish time, etc.

Analytics that are generated by Accern's analytics engines using the metadata information is classified as derived analytics. Below we present the list of analytics present on the output data.

Definition: Uniquely identifiable ID that is randomly generated.

Process: Each signal is a theme extracted from the document. While creating a signal, a random unique identifier function is used to create the ID that is used as a string value.

Definition: Unique ID assigned to the document.

Process: For every document that is processed (news, blog, etc.), we have a unique ID that helps us identify specific articles. One document can have multiple signals associated with it.

Definition: The source of the document.

Process: We look up a parent source of information. If the domain information is not available, the doc_source is extracted from the unique URL of the article. For the local database, the doc_source information is either set to 'custom' or any other category value provided to us.

Definition: Classifies document based on where the article/document is published and its mode of access.

Process: To determine the doc_type, we first look at the mode through which we found the document. For instance, if a document is found in one of the RSS feeds, then the doc_type will be set to 'Feed.' If it was accessed through a premium news feed API, then the value is set to 'Premium News'; or when it was accessed via SEC's EDGAR database, the value is set to 'SEC Filing,' and so on. Next, we use the doc_source information to classify whether the document is news or blog. Accern actively maintains a mapping between doc_source by doc_type, like news/blogs. When a new source is encountered for which there's no historical data or news/blog mapping, the default value is set to 'blog.' The Accern team then reviews these sources and the mapping files are updated periodically.

Definition: Content of the original document. This field is only available in particular scenarios.

Process: The original text of the document as scraped/extracted from the source.

Definition: When the original document was crawled (at time).

Process: This info is generated when the document is actually crawled.

Definition: The time when the doc was published.

Process: If the published date is extractable from the document, it’s used; otherwise, it’s the timestamp of when the document was crawled or provided to Accern.

Definition: ID of the data provider.

Process: N/A

Definition: The title of the original document.

Process: N/A

Definition: Online URL for the original document.

Process: The original URL of the document as extracted from the source.

Definition: Document ID given by a third-party data provider.

Process: ID of the document that was assigned by the original source.

Definition: The average sentiment of all entities and events in a document.

Process: A simple averaging of all the signal_sentiment in a document, which in turn are overall signal sentiments.

Definition: It is a unique identifier for clusters to which a given article belongs. By tracking a doc_cluster_id, a user can trace how different articles evolved.

Process: We group similar documents into clusters based on events and entities discussed in the document.

Definition: It is a theme identifier.

Process: It is formed by the combination of the entity_accern_id and event_accern_id.

Definition: Overall relevance of the signal.

Process: It is calculated as the average of entity and event relevance. The entity and event mentioned are the ones present in the signal itself.

Definition: Overall sentiment of the signal.

Process: It is calculated as the average of entity and event sentiment. The entity and event mentioned are the ones present in the signal itself.

Definition: A boolean indicator for whether a signal is one of the most relevant signals of the document or not.

Process: Max signal relevance is calculated for each document. The signal(s) with the signal relevance equal to max signal relevance are classified as primary signals.

Definition: Accern id of the entity

Process: Entities are extracted by Accern's proprietary entity extraction models. Once an entity is extracted from a theme, we retrieve its Accern ID available in our databases and update the entity_accern_id value.

Definition: Scores an entity based on the emphasis it receives in the document.

Process: To determine the relevance of an entity, we consider the following two factors - a) the number of times an entity is mentioned in the article and b) the positions within the text where the entity is mentioned. We then combine these two factors into a single relevance score. Entities mentioned frequently and appear earlier in the document receive higher relevance scores than entities mentioned relatively fewer times and mostly in the later sections of the document. It is important to note that a document may have multiple highly-relevant entities. On the contrary, we reject any document that does not contain any relevant entity.

Definition: This is the sentiment value calculated for each entity based on the text surrounding it.

Process: We calculate the sentiment using Accern’s proprietary sentiment analysis models.

Definition: Identifies which knowledge graph was used for this entity (primary or custom).

Process: Entities in our databases are classified as ‘primary’ and the ones added by the client are classified as ‘custom.’

Definition: It is the traded ticker symbols of the extracted entity.

Process: Entities are extracted by Accern's proprietary entity extraction models. Once an entity is extracted from a theme, we retrieve its traded symbol available in our databases and update the entity_ticker value. Our ticker symbol database gets updated every night with information such as initial public offerings (IPOs), symbol changes, sector/industry updates, entity name updates etc.

Definition: It is the stock exchange code where the entity is traded.

Process: We refer to the 'Entities' database and update the entity_exch_code field.

Definition: It is the name of the company as it is registered on the stock exchange.

Process: Accern has access to the list of all companies traded on each stock exchange. We actively maintain this 'Entities' database with important corporate events that may affect an entity's name, sector, stock ticker, etc.

Definition: It is the type of entity, such as public equity, commodity, cryptocurrency, etc.

Process: We refer to the 'Entities' database and update the entity_type field.

Definition: A list of popular indices where the entity is a constituent.

Process: We refer to the 'Entities' database and update the entity_indices field.

Definition: Figi Code of the entity (asset class).

Process: We refer to the 'Entities' database and update the entity_figi field.

Definition: It is the parent country of the entity.

Process: We refer to the 'Entities' database and update the entity_country field.

Definition: Share class Figi code for an entity (asset class).

Process: We refer to the 'Entities' database and update the entity_share_class field.

Definition: It is the region where the entity is traded.

Process: We refer to the 'Entities' database and update the entity_region field.

Definition: It is the sector the entity belongs to.

Process: We refer to the 'Entities' database and update the entity_sector field.

Definition: Hit word(s) of the entity.

Process: A list of words is generated for the entity hits by the Accern proprietary API. A distinct list of hits is then extracted for the entity_hits field.

Definition: Text surrounding the tagged entity.

Process: Accern’s proprietary API recognizes the relevant text surrounding the tagged entity in order to update the entity_text field.

Definition: Additional information associated with the entity.

Process: We refer to the 'Entities' database and update the entity_attributes field.

Definition: Accern ID of the event

Process: Events are extracted by Accern's proprietary event extraction models. Once an event is extracted from a theme, we retrieve its Accern ID available in our databases and update the event_accern_id value.

Definition: Scores an event based on the emphasis with which it is mentioned in a document.

Process: To determine the relevance of an event, we consider the following two factors - a) the number of times an event is mentioned in the article, and b) the positions within the text where the event is mentioned. We then combine these two factors into a single relevance score. events that are mentioned frequently and appear earlier in the document receive higher relevance scores than events that are mentioned relatively fewer times and mostly in the later sections of the document. It is important to note that there may be multiple highly-relevant events in a document. On the contrary, we reject any document that does not contain any relevant event.

Definition: This is the sentiment value calculated for each event based on the text surrounding it.

Process: We calculate the sentiment using Accern’s proprietary sentiment analysis models.

Definition: Event groups are the broader financial events category that contains multiple related events.

Process: Accern has developed a financial event tree that contains over 25+ financial event groups, 250+ financial events, and over a million financial phrases. A financial event can only be part of one event_group, whereas, each event_group can contain multiple financial events. Once a financial event is extracted by the event extraction model, we search for the parent group in our database and update the event_group field.

Definition: Financial events extracted from the stories.

Process: We actively maintain this 'Events' database that contains important corporate events. Each signal contains a unique financial event for a specific company.

Definition: Text (words/phrases) as the event was found in the document.

Process: A list of words is generated for the event hits by the Accern proprietary API. A distinct list of hits is then extracted for the event_hits field.

Definition: Text surrounding the tagged event.

Process: Accern’s proprietary API recognizes the relevant text surrounding the tagged event in order to update the event_text field.

Definition: Additional information associated with the event.

Process: We refer to the 'Events' database and update the entity_attributes field.