This functionality is implemented using QMiner data analytics platform and is opened to other solutions. Within the toolkit, QMiner provides the functionality of learning models from historic datasets and use them for prediction on new data points. It implements a comprehensive set of techniques for supervised, unsupervised, and active learning which support bot structured and unstructured data. The figure below presents an overview of QMiner architecture.
A specialised Toolset for Event and Weather Analytics was built on top of QMiner to streamline the work with these specific types of data. It contains all the functionality needed to ingest data for analysis, process the data to produce predictive models and then use these models to produce predictions. Data ingestion is possible from different sources and formats including csv files and SQL as well as non-SQL databases. For the modelling step the entirety of the QMiner modelling library is available. Finally, a simple and efficient REST server is available for querying the models and obtaining predictions. The toolset is optimised for tasks such as predicting sales spikes based on historical data and contextual features such as the example in the figure below.
The Keyword Clustering Tool clusters together keywords by their meaning, producing sets of keywords related to some central keyword (i.e. a category name). This solves a common problem in data organisation such as categorising novel marketing keywords for online campaign optimisation. It transforms the keywords into a low-dimensional semantic space where they are comparable among each other. It uses the FastText text representation library and through it supports 157 languages. The transformation is created in such a way so that semantically related words are closer together than those that are unrelated which enables clustering using standard machine learning methodology (see illustration below).
Back to top