Rob van Zoest

Innerdoc builds custom Business Solutions on textual data from your domain. Because your business has specific needs around Text Mining and Natural Language Processing. Whether it's cloud-based or on-premise, ad-hoc or automated. We build our own content-parsers, especially in Dutch and English.

This demo is created to show the integration of Streamlit with TimelineJS. It became a timeline about the history of Natural Language…

It started with the idea for a fun pre-christmas project and with the help of Plutchik’s Wheel of Emotions and Zero-shot emotion…

A project to build a mini guide with more than 80 helpful NLP task descriptions! Structured in the format of the Periodic Table.

An overview of all demo's. Interaction with a demo beats reading and telling.

Why compute Affective Meaning? Applications can be Detecting: sentiment towards politicians, products, countries, ideas frustration of…

Sentiment Analysis can be considered a subfield of information extraction. It is used in a wide range of areas and is sometimes referred to…

Sometimes the problems encountered when trying to understand a text, or better, a corpus of documents becomes so complex that you need to…

Visualizing a network of interlinked descriptions of entities.

Plotting geocoded Named Entities on a map.

Increasing insight into text by plotting events chronological on a timeline.

Visualizing Word Embeddings is often done to inspect the embedding and experience the cohesiveness of a subset of the embedding.

Most Wordclouds seem to ignore best practices for visualizing information.

Printing text, but prettier. Text is not only characters, but also meta-info.

Presenting your NLP task results in a transparent, interactive and fancy App.

Large scale content analysis for identifying, collecting and producing information to support investigations and understand the voice of the author.

Discovering facts about entities (NER, NEL) and building a knowledge base with it.

Increasing search accuracy by adding semantic information about a piece of text to the index, in addition to keyword search.

Programming a natural and convincing chatbot dialogue for the personas of your customers, to meet your customers’ needs.

Answering questions posed by humans in a natural language, regardless of the format of the question.

Extracting semantic relationships from a text, to make a connection between entities.

Generating long sequences of words, based on long sequences of words with appropriate performance.

Expressing the meaning of a source text into a new text by using different words and maintaining the semantic meaning.

Abstractive summarization systems generate new phrases that express a text by using as few words as possible.

Translation by machines to perfectly transform text into another language.

Writing sentences based on structured data is also called Data-to-Text Generation.

Predicting the next word that is appropriate in the context.

Word Representations with the ability to incorporate context.

Multi-dimensional meaning representations of a word are reduced to a level of N dimensions, so the vectors can be used for similarity measures.

Estimating the degree of similarity between the semantic representation of two documents.

Measuring the syntax similarity or semantic word similarity by a specific distance calculation.

Defining lexical databases which consist of concepts that are described and interlinked by means of conceptual-semantic and lexical relations.

Finding text that is exceptionally far from the mainstream text.

Quantifying the deviation of the occurrence of words beyond the expected variability, and defining above what threshold you call this a trend.

Dividing a set of vectorized documents into N unsupervised topics by determining how similar vectors for a specific topic should be, and how many topics should be distinguished.

Extracting the most relevant sentences from a text works in the same way as Keyword extraction.

Providing the most relevant words from a document.

A specific sub-solution of Text Classification is Multi-Label Multi-Class Text Classification.

Assigning tags or categories to text according to its content. It is the broader task where Intent, Sentiment and Spam classification are part of.

Understanding the user’s intent and giving the correct responses.

Detecting the overall attitude expressed within a text in an imperative need to standardize the measurement of human sentiment and affective meaning.

ISPs continuously need to improve on detecting and filtering spam out.

Monitoring your Language Model might give you the feedback to further improve on performance and usage.

Deploying your Language Model might be a recurrent building block fro DevOps in a larger pipeline.

Explaining the outcomes of your Language Model is needed to prevent distrust and increase transparency.

Evaluating the quality of a Language Model should be done by comparisons based on the right metrics for your model type.

Training Language Models should start with a simple baseline and be improved with more complex techniques.

Identifying the language of a text is often done before you select the right language model.

Extracting text from a file should be accompanied with the extraction of meta-information.

Pre-processing text with the goal to increase the quality of subsequent NLP tasks.

Finding texts that are exactly the same or show a high similarity. Similarity can be measured on lexicality or semantic meaning from embeddings.

Measuring the readability of a text by looking at the keyword density, syllable count and the average length of sentences and words in a document.

Improving the grammar on a sentence level.

Splitting text into paragraphs requires more custom logic. A paragraph might contain a more comprehensive meaning than a sentence.

Finding the words that together form a sentence, or from another viewpoint, detecting sentence boundaries.

Removing sensitive information before a document is shared with others. Deidentification and obfuscation of persons and organizations relies on Named Entity Recognition.

Finding all expressions that refer to the same entity in a text. You can compare this to Named Entity Linking, but it doesn’t necessarily use a knowledge base.

Assigning a unique identity from a knowledge base to a named entity.

Finding strings that contains an indication of time and then extracting a normalized time format out of it.

Parsing text into an address and converting an addresses into geographic coordinates like latitude and longitude.

Extracting price and currency from raw text and normalize it into a standard format.

Abbreviations are an efficient way of writing, but it lowers text comprehension. To solve this, identify the long-form to enrich the short-form.

Identifying named entities is the task of assigning a NER category, like Persons, Locations or Organizations, to words in a sentence.

Breaking text into verb- or noun-phrases result into semantically correct subphrases of a sentence that are deducted from the dependency structure.

Finite size lookup tables might inspire your to build rulebased searches, especially when semantic info like lemma’s and POS- and Dependency tags can be used in the search pattern.

Detecting N-grams results in common multi-word expression with a high probability of occurrence, like the Bi-gram ‘red wine’.

Ignoring the meaning of a negation will flip the polarity of your text.

Spell Checkers can recommend corrections on the word level.

Besides Stemming or Lemmatizing, there still might be a need to edit words to move to more default words.

Lemmatization usually refers to rewriting a word to its base form (lemma) properly.

Stemming refers to a crude heuristic process that chops off the ends of words in the hope that words with the same meaning become words with the same syntax.

A Dependency Parser extracts a dependency graph from a sentence. In the graph the grammatical structure, like subject and object, and relationships between words represented by Dependency tags.

The syntactic function of a word, like Noun or Verb, is defined by the Part-of-Speech (POS tags) and is based on the context.

Assigning additional morphological information clarifies the grammatical meaning of a word, additionally to the syntax.

The goal of tokenization is a vocabulary. Word-based tokenizers have larger vocabularies with more Out-Of-Vocabulary (OOV) words than sub-word vocabularies.

Many NLP tasks make use of intensive matrix calculations, for which word id’s are used, rather than words. For this, raw text is split up into tokens that represent (sub)words.

Programmatically build training datasets by defining heuristic rules which are used in functions for labeling training data.

Boost your performance by creating data out of data, instead of new data.

Creating training data is a labor-intensive task. Fine-tune the training data definition yourself and then scale-up by outsourcing to remote workers.

Gold data contains the ground truth. Re-use available resources, but be careful that the dataset matches your purpose.

Use an annotation tool that benefits from active learning to enforce a robust annotion process and balanced annotations.

Nobody wants to do the manual labor of tagging. Everybody wants to build language models with annotated training data.

Extracting text and transforming it to qualitative data is challenging when you have the output format (PDF or image) and need to reduce it to the source format of textual data.

The lack of a Corpus or API requires you to scrape your textual data or files from the web. Overcome the challenges of IP-blocking, cookie walls, request headers and js-websites.

An API serves as the interface between different applications. The requestor automatically gets access to data, with the benefit that the source doesn’t have to know how the other system exactly works. 

A Corpus is a language resource consisting of a structured set of documents and additional information. It incorporates pre-processed documents and their meta-data, which might be the output of other NLP tasks. 

Structured data implies ready-to-use data, but you have to interpret it’s scheme.

Typing your own text gives you more confidence when testing your code or a demo.

Text is made of characters, but files are made of bytes. These bytes represent characters according to some encoding (aka character set). Fix or load your data by choosing the right encoding.

Innerdoc builds custom solutions with Text Mining and Natural Language Processing techniques. Default language models are of no use, because they don't recognize the specific language from your business. Whether it's cloud-based or on-premise, ad-hoc or automated. We build our own content-parsers, especially in Dutch and English.