How Semantic Analysis Works

Learn how Workplace Dynamics analyzes collaboration data and draws meaning from the text.

Workplace Dynamics uses Natural Language Processing (NLP) to understand, interpret, and extract meaningful insights from emails, messages, reactions, and other data generated by the organization's digital communication tools. NLP is used to recognize the emotional tone and the presence of tasks or conflict in communications between employees. Semantic analysis identifies the causes of employee burnout and potential solutions to help employees cope with the stress.

Machine learning is used to process information, including text that is presented in other languages. A key tool for analyzing natural language text is a family of deep neural network-based classifiers. The algorithms are trained using examples of text that are labelled with a classifier (category). The algorithm learns the key features and structure present in the texts that are associated with a specific classifier. Once trained, the machine learning model can predict the category that any given text can fall into.

How is insight extracted from the text?

We understand that messages can contain extremely sensitive information, so we use a technique that assigns predefined categories to the text. Classification is very similar to how email systems use machine learning to detect unwanted messages (spam). After the contents of an email is analyzed for spam, the system adds a marker to the email header to indicate the nature of the content without being specific. The marker does not tell you anything about the information that is contained in the email such as what kind of product is being advertised, how much it costs, and whether it is advertising at all. After the collaboration data is analyzed, the messages are categorized with a set of markers that allow us to make general conclusions about the information contained in the text.

Messages are marked using the following categories:

• Conflict
• Praise
• Sentiment
• Task

Conflict

Text is analyzed to determine if it contains the presence of conflict. Does the text contain disagreement, missed deadlines, or information that may adversely impact the organization. Conflict can be very diverse in nature. Conflict can be interpersonal when two people are in a state of hostility. It can also occur in groups, for example, when different departments cannot align on responsibilities. In text, conflict can be expressed in a non-trivial way, but it is important to separate conflict from sarcasm and jokes. So how do we avoid mixing up conflict and sarcasm? Neural networks and their capabilities are a reflection of the data on which they were trained. We train the neural networks using large volumes of texts, which are manually marked by experts. We ensure that the content of the training texts reflect all aspects of human communication as fully as possible including humor and sarcasm.

Praise

Text is analyzed to determine if it contains gratitude and compliments. Praise is a complex action and is a significant marker of a healthy team environment. It is considered positive and serves as an assessment of achievements, a powerful motivational driver, and a form of feedback at the same time. Praise can be expressed to a group of employees or the whole organization.

Sentiment

Text is analyzed for its emotional tone to determine if it is positive, negative, or neutral. Only the informative part of messages are analyzed so greetings and courtesies are not considered as factors that affect the positive or negative sentiment of the message. If there is both negative and positive in one sentence, the entire sentence is considered as negative. Message sentiment is an indicator of the health of an organization.

Task

Text is analyzed to determine if there are direct instructions for a specific person or group of employees.

Frequently Asked Questions

Does Workplace Dynamics store any collaboration data?

Workplace Dynamics does not store email and message content. We do not provide access to the collaboration data to anyone. Instead, a special semantic vector is stored, which cannot be used to restore the text, numbers, names, and other factual information contained in the original message.

What languages were used to train the algorithm?

The NLP supports more than 90 languages for semantic analyses. The algorithm was trained on the following languages:

Afrikaans, Albanian, Amharic, Arabic, Armenian, Aymara, Azerbaijani, Basque, Belarusian, Bengali, Berber languages, Bosnian, Breton, Bulgarian, Burmese, Catalan, Central/Kadazan Dusun, Central Khmer, Chavacano, Chinese, Coastal Kadazan, Cornish, Croatian, Czech, Danish, Dutch, Eastern Mari, English, Esperanto, Estonian, Finnish, French, Galician, Georgian, German, Greek, Hausa, Hebrew, Hindi, Hungarian, Icelandic, Ido, Indonesian, Interlingua, Interlingue, Irish, Italian, Japanese, Kabyle, Kazakh, Korean, Kurdish, Latvian, Latin, Lingua Franca Nova, Lithuanian, Low German/Saxon, Macedonian, Malagasy, Malay, Malayalam, Maldivian (Divehi), Marathi, Norwegian (Bokmål), Occitan, Persian (Farsi), Polish, Portuguese, Romanian, Russian, Serbian, Sindhi, Sinhala, Slovak, Slovenian, Somali, Spanish, Swahili, Swedish, Tagalog, Tajik, Tamil, Tatar, Telugu, Thai, Turkish, Uighur, Ukrainian, Urdu, Uzbek, Vietnamese, Wu Chinese, and Yue Chinese.

The model also works well with minority languages and dialects such as:

Asturian, Egyptian Arabic, Faroese, Kashubian, North Moluccan Malay, Nynorsk Norwegian, Piedmontese, Sorbian, Swabian, Swiss German, and Western Frisian.