In any language processing (NLP) system, some words and phrases can create tricky precision or accuracy issues in the output. In these cases, the Lexalytics professional services team works with our customers to solve your problems quickly and effectively.
First, we evaluate whether or not we can fix your problems by tuning the underlying text analytics. In essence, tuning means telling the system exactly what to do. Tuning is immediate, precise and transparent.
Our Methodology
- First, tune your system as much as possible.
- If needed, train the smallest feasible machine learning model.
But tuning is less flexible than machine learning. This means that sometimes it will be more efficient to solve your problem by training a machine learning model. To reduce costs, risks and training time, we build small, targeted models to solve one specific problem at a time.
Tuning and Configuration Services
Entity Configuration
We’ll build and configure entity lists tailored to your industry, including products, brands, or even addresses.
Custom Categorization
We can improve the accuracy of your categorization and topic extraction by building custom taxonomies to accurately sort content by specific aspects, including types, features, or characteristics.
Sentiment Tuning
When particular words and phrases carry a unique sentiment weight in your business or industry, we can tune our sentiment scoring systems to reflect your perspective.
Custom Machine Learning Models
Sometimes, an accuracy or precision problem is too large or unwieldy to solve with tuning alone. In these cases, we use machine learning models for a more cost-effective solution.
To reduce your cost and risk, Lexalytics trains machine learning “micromodels” to solve very specific problems, such as entity recognition of a single ambiguous company name or categorization of food products and sauces.
Micromodels
- Require less data
- Are easier to grasp and debug
- Have fewer unwanted side-effects
We also build models to extract non-traditional data, such as stock ticker symbols, medical billing and treatment codes, age ranges and deadlines. Contact us to discuss how machine learning can help solve your data-related problems.
Machine Learning Solutions
Suggesting Answers to Medical Inquiries
Contact center operators at Biogen’s medical information department used to comb through large resource libraries to find answers to incoming inquiries. Lexalytics trained machine learning models to suggest answers instead, as part of a larger solution to accelerate their operations.
Sentiment Scoring of Context-Dependent Phrases
The phrase “wicked sick” should receive a negative sentiment score in the context of healthcare. But the same words can be neutral or even positive in the context of video-gaming or sports. We trained a machine learning model to accurately score the sentiment of this phrase based on the context in which it appears.
Entity Recognition on Ambiguous Company Names
“ Apple” can represent a fruit, a company, or even an adjective (apple-bottomed jeans). For a financial services customer, we trained a series of machine learning models to improve the reliability of entity recognition on ambiguous words like this, even when they’re misspelled or abbreviated.
Services Summary
Lexalytics is the industry leader in transforming unstructured text into usable data and insights. Our professional services team helps our clients gain the most possible value from their text analytics solutions.
| Service Type | Details | Duration |
|---|---|---|
| End-to-End Processing | Just send us your data and let us do the rest. We’ll provide you with actionable business insights presented in clear visualizations, ready for delivery to your clients or co-workers. | varies |
| Custom Categorization | We can help you create and manage a variety of tailor-made categorization methods: custom codeframes, unique taxonomies, and complex categorization schemas. | varies |
| Sentiment Tuning | Identify and optimize specific sentiment phrases for your particular business and verticals | 5d |
| Entity Tuning | Define and analyze unique entities based on your organization’s requirements | 5d |
| Platform Migrations | We’ll help you seamlessly transition your existing text analytics software, including custom queries, sentiment phrases, and named entities, into our software platforms | 5d |
| Custom Machine-Learning Models | We take your annotated content and train a machine learning model on it, picking the best fit model depending on whether you favor precision, recall, or a blend.Deploy the models in Semantria, or give you the model files for your Salience deployment. | 2d |
| Accuracy Evaluation | Train your data science team on best practices for evaluation precision and recall on a variety of text analytics functions such as sentiment, named entity extraction, and classification.Perform a sample assessment of Semantria or Salience against a set of “gold standard” content you provide | 1d |
| Salience Integration | Train your development team on the Salience API including basic functionality, performance characteristics, and best practicesTrain your development team on how to configure Salience NLP extractionDesign an integration scenario that fits your current content pipelineDeal with multi-tenancy (configuring one text mining system to serve many clients with different needs)Write sample code in the wrapper of your choice to process samples of your content | 3d |
| Semantria Integration | Train your development team on the Semantria API, including basic functionality, configuration management, and content processing best practicesTrain your development team on how to configure the Semantria extraction rulesDesign an integration scenario that fits your current content pipelineDeal with multi-tenancy (configuring one text mining system to serve many clients with different needs)Demonstrate use of our Semantria SDKs to speed development | 2d |
| Title | Audience | Duration |
|---|---|---|
| Semantria / Developer introduction | Developer | 1d |
| Salience / Developer introduction | Developer | 1d |
| Tools / How to use our apps | Data analyst | 1d |
| Sentiment / Evaluate and tune sentiment | Data analyst | 1d |
| Classification / Taxonomy definition and query building | Data analyst | 1d |