Approach

Together with our clients, we build solutions which are bridging the gap between NLP technologies and business problems. Two approaches exist: the deepset Spotlight generates relevant insights from existing data and provides a basis for strategic and tactical decisions. The deepset Engine covers the life cycle, from ideation to production, of tailored solutions which enhance operational excellence. Whatever process is pursued, the objective is the same - delivering value.

deepset Spotlight
In a deepset Spotlight we empower our clients to make enhanced decisions based on existing unstructured text data. Various methods are used to generate a sound reporting on relevant strategic and tactical business challenges. Product decisions, segmentations or necessary process redesigns are exemplary implications our clients successfully implemented after working with our insights. 

Discovery

Discovery

Together with our clients we review problems, desired insights and get an overview over existing data sources. We use the Discovery to understand the business and to assess which insights we can generate. Finally, we use the Discovery to agree on a project plan and delivery milestones.
  
 
Data Analysis

Data Analysis

Having a clear objective in mind, we work with provided data. No matter if E-mails, newspapers, voice recordings - we prepare the data and apply various methods on it. Constantly, we are in close contact with our clients - working with data has always explorative elements and additional insights might come up while working with the data. The results are consolidated in a report and raw analysis is provided in the format which supports our customers decisions best.
 
Implication Analysis

Implication Analysis

We are not just experienced in working with methods of NLP - our team always bridged the gap from technology to applications. Taking our experience from prior projects we contribute in the definition and valuation of implications on the business of our clients. Many insights also provide perspectives for the productive use of NLP - many of our clients moved from the Spotlight to an Engine, solving major strategic issues on an operative level. Our goal is to accompany you, no matter which implication is drawn.
deepset Engine
The deepset Engine is our approach to tailor solutions to our clients' issues and their system landscape in order to gain value through improving their own operations. Systematically, we identify and qualify the opportunities of NLP and even operate the solutions when implementation is done.
 
Use Case Discovery

Use Case Discovery

Together with our clients we discover the potential for using NLP to unlock value and solve business and process challenges. This is done by listening to our clients issues but also by understanding their business - the opportunities are larger than many clients think and it is our mission to broaden their mind and open new fields of applications for the technologies in our portfolio. The outcome of the discovery are different use cases as well as an agreed project plan with deliverable milestones.
 
 
Proof of Concept

Proof of Concept

With clear use cases, the work with data starts. We use this phase to tailor the technologies and prove the value in real settings. Besides the functional proof of concept, we start to collect our clients' requirements on a productive system - this includes especially information on the existing IT landscape and systems. Besides the prototyped solution, the specification for the productive system is one major deliverable of this phase.
 
 
Productization

Productization

From an idea to the prototype and finally to the product - this phase is used to develop a running system, fulfilling all requirements regarding scalability, availabiliity and integrations with existing system. The core of this phase is the implementation, testing and finally the deployment of the solution.
 
 
Operation & Maintenance

Operation & Maintenance

The last phase of the solution life cycle is the most exciting one for our clients - it is the phase in which the value is delivered on a daily basis. We ensure that our clients can focus on their business. We take care about running, availability and continous updates - technology changes fast and we use this phase to keep our clients on track. Besides technological change, we also consider every change made in integrated  systems. We use this phase also to engage with our clients ongoingly - to see what can be improved and also which other opportunities arise. Not just the machines are learning - but so do we.

Technologies

Document Classification

Automatically detect and tag different types of documents.

Doc2Data

Extract structured data from your documents with customized entity recognition.

Sentiment

Detect emotions in your texts and conversations.

Voice2Text

Create new opportunities for spoken language by transforming it to text.

Summarization

Condense texts to their essence.

Topic Extraction

Learn the key topics across a large corpora of documents.
documentclassification_adjusted.jpg

Document Classification

Information in today's businesses is often buried in unstructured text documents such as invoices, analyst reports or customer emails. The automatic classification of documents into fine granular categories (e.g. customer emails into request types) is often key to further process automation.

Deep neural networks have reached human-level accuracy in many of such document classifciation tasks. The models are often trained in a two-step procedure: Teaching the general properties of the language and then training it for the specific classification problem using many examples. While we have models who master the general properties of language, the second step is completely tailored to your individual case.

Sentiment Analysis

Understanding the emotions within communication is crucial for human interactions and the correct interpretation of information. While this is easy for humans, it used to be tough for computers - until now. 

The machine learning models are usually trained on large external data sets and can rapidly be adapted to your specific use case. Whereas the classical approach distinguishes between positive and negative emotions of a text, recent research explores more granular emotional dimensions and the mapping of emotions to entities (e.g. products).
sentiment_adjusted.jpg
summary_adjusted.jpg

Summarization

Texts and conversations are full of unnecessary details. With the help of summarization models texts can be condensed to their essential information.

Models are distinguished into two types: extractive and abstractive. Whereas the first type uses sentences from the original text for the summary, the latter generates completely new sentences. Models are trained on large external data sets and finetuned for your specific domain.

Doc2Data

Documents often contain valuable information, but further processing in a structured manner is difficult without human intervention. Models for Named-Entity Recognition (NER) allow the automated extraction of predefined entity types like companies, persons or locations. The extracted entities can be passed to other systems (e.g. CRM / ERP) or used for further analysis.

Deep neural networks or Conditional Random Fields are the dominant methods in this area. Entity types are not limited to the general ones mentioned above. They can be customized to individual interests (e.g. your product names or fields in a form) as long as annotated training samples are available.
doc2data_adjusted.jpg
voice2text_adjusted.jpg

Voice2Text

Speech recognition is one of the oldest challenges in machine learning, but only recently models reached levels of accuracy that are finally on par with humans. This created opportunities for long envisioned consumer devices like Alexa, but also enables innovative business applications. For example, converting customer calls to text opens up a new spectrum of analytical opportunities. 

Leveraging the models from the best players in the market, we optimize the performance for your domain. Depending on your use case this can include custom preprocessing of audio, model adjustments or postprocessing of recognised text. For example, we can hide confidential information like credit card details or add custom vocabulary from your domain. 

Topic Extraction

Topic models infer abstract "topics" that occur in a collection of documents and can be used to explore, understand and cluster a large collection of documents. For example, analyzing the topics across customer requests helps to understand repetitive topics not covered by your FAQ. 

By modeling hidden semantic structures in a text body the algorithm detects the major topics across all documents. Each document can then be described as a distribution of these topics and each topic is described as a distribution of words. Since the models are trained in an unsupervised manner, there's no not need to have any labeled training examples.
topic_extraction_adjusted.jpg
©2018 deepset. All Rights Reserved.
We use cookies to provide you the best experience on our website. By continuing to browse this site you agree to the usage of cookies.
More information