Anno 2021, companies in retail have more and more data at their disposal. Data about their business processes, customers, products; virtually every aspect of modern business is subject to large amounts of back-end data. This data can then be used to make rational and informed decisions and strategic decisions. Furthermore, this data can be used, for example, to optimize the sales process or to provide customers with better service. In short, data is invaluable to companies that want to move with a digital paradigm that is constantly shifting.
However, just the presence and quantity of data is not enough to speak of Data driven value creation, the quality of the present data is equally important as incomplete data or poor quality data will only lead to incongruent conclusions and unsubtle decisions. This blog will therefore elaborate on the extraction of data and how data enrichment benefits the quality – and thus the use – of data.

Product Named Entity Recognition

Product Named Entity Recognition’ better known as; P-NER is a method for extracting information from large, unstructured, and textual data. P-NER manages to classify data into categories that need to be defined in advance. For example, an example to illustrate the use of P-NER might be related to products that consist of multiple, distinct properties. For example, a television has a brand, size, weight, resolution, and so on. These properties are then classified under the corresponding categories that are predefined. However, P-NER still requires traditional machine learning methods and significant human contribution which is obviously not desirable. Deep learning could be the solution to this in two different ways explained below.

Hybrid Bidirectional Long Short-Term Memory.

‘Hybrid Bidirectional Long Short-Term Memory’ – in short; BI-LSTM – is a P-NER application that consists of three different layers, namely: ‘input representation’, ‘context decoder’, and ‘tag decoder’. The first layer helps a model to understand data and interpret it properly, the second layer enables processing of images by ‘unfolding’ the input, so to speak, into different, underlying structures and properties. The last layer does the same as the second layer but for textual input.


BERT (Bidirectional Encoder Representations from Transformers) is a language model that manages to understand and contextualize texts in such a way that, based on connections made between words, certain words can be assigned a value. Following on from Feature extraction, this value may include, for example, recognizing a word as a feature. In the sentence; “The car is sprayed in a blue hue that reminds you of the Azure.” – ‘Sprayed’, ‘azure’, and ‘tint’ all say something about the word ‘blue’, using training data BERT can relate these words to each other and thus recognize the color ‘blue’ as a feature. This very simple example shows how a language model like BERT can be used to enable the extraction of product properties from unstructured text. The more training data a model like BERT has at its disposal, the more accurate this method of Feature extraction is.

Data Enrichment

As mentioned earlier, good quality data must underlie the extraction of product attributes in order to be able to speak of Data driven value creation. Data can of course be enriched manually, but this is a time-consuming task and also prone to human error. To avoid this, data needs to be checked for inconsistencies and other faulty properties after it has been enriched, again an intensive task that puts the entire process behind the manual enrichment of data into a downward spiral of inefficiency in terms of time, resources, and costs.

PowerEnrich as a software engages both the extraction and enrichment of data to enable a comprehensive approach on Data driven value creation in an autonomous and simple manner. PowerEnrich enables extraction of data from four different types of sources, namely images, text, PDFs, and web pages. Thanks to smart use of AI, PowerEnrich software manages to recognize and understand data regardless of the use of abbreviations, different spellings, or expressions.

In short, PowerEnrich succeeds in helping companies process their product data and product attributes faster and better. In addition, PowerEnrich enables more detailed and comprehensive product descriptions resulting in better product findability, increased sales and a better customer experience.
Curious about how PowerEnrich can help your company? Please contact us to explore the possibilities together.

By Lieske Trommelen