All types of textual data are compatible with Proxem Studio: plain text, web pages, emails, PDF, Word documents, etc. Depending on requirements, pre-processing integrated into Proxem Studio can be applied to the data, such as cleaning up HTML tags for pages or emails.
Proxem Studio can integrate millions of verbatim without difficulty. For the automatic suggestion algorithms based on machine learning to work perfectly, it is recommended to have at least 2,000 examples as input. However, Proxem Studio also works on low volumes thanks to our linguistic resources and the powerful semantic rule engine.
Proxem Studio integrates a web page collection module for uses such as market mapping or technology watch.
Proxem does not offer a social media collection solution, but it is possible to connect your favorite social-monitoring software with Proxem Studio to analyze verbatim from the social web. It is also possible to use the Google My Business APIs to integrate data from Google Avis. Proxem Studio is natively connected to many feedback management solutions such as Critizr, Mediatech-CX, WizVille, Satisfactory, and more.
Data security and confidentiality is a major concern at Proxem. In general, we do not recommend integrating personal data in Proxem Studio. In most cases, we do not need to have the information identifying the author of a verbatim or a document. Proxem will advise you in setting up your project to protect personal data present in the SaaS platform as recommended by EU regulation n° 2016/679 (known as "GDPR"), in particular via three areas of work:
By including only the metadata necessary for analysis needs.
By promoting the use of keys that only the customer can associate with the person concerned.
By applying pseudonymization on text content (optional): Proxem Studio makes it possible to anonymize data by removing collected verbatims from information such as name, email, address, telephone number...
The reliability of the analysis depends on several parameters, and the objective that we set. The two main measures that we will seek to optimize in semantic analysis are:
The reminder: does the system classify all the verbatims or all the documents? Are all opinions well categorized as positive, negative or neutral? In general, recall is very difficult to measure in absolute terms, since it would be necessary at the start of all the data classified by a human to know if the system "forgets" something. One way to resolve this difficulty is to approximate the coverage rate: how many of my documents are automatically classified? We can have 100% reminder by classifying all documents in all categories, but we cannot say that in this case the system is relevant...this is why we must combine the reminder with precision.
Precision: when the system classifies, does it do it correctly? Precision is generally measured by sampling: we export a subset of the results of the analysis which we submit to a human expert who will judge on each decision of the machine whether it is correct or not.
Depending on the needs of the project, one of the metrics will be favored.
The combination of recall and precision gives an overall analysis performance score, the F-measure, which is expressed as a percentage. In the best cases, we have obtained scores of up to 95% F-measure. But we must keep in mind that it remains dependent on the sources, the number of classes or concepts to be detected in the text, and... the budget.
The quality of our deliverables is our Number 1 priority and Proxem Studio includes a measurement and monitoring module for the quality of semantic analysis to allow monitoring and continuous improvement throughout your projects.
More seriously, irony or sarcasm are very difficult to detect in general: is a verbatim that says "bravo Macron" sarcastic or is it genuine?
Proxem Studio allows, in some cases, to understand the irony but on the condition of having more information in general than the verbatim alone. Typically, in an NPS project where the score is 0 and the verbatim "thank you for the delay", the reason detected will be a criticism of the delay, the thank you being detected as ironic and not congratulations.
The classification plan is evolving in Proxem Studio: we support you on a daily basis to bring it to life according to your feedback, new themes that appear and new needs.
More generally, it is also common to have several filing plans according to the objectives of different departments in your organizations. In addition to the verbatim classification plan, Proxem Studio also provides a complete thesaurus of the themes addressed, allowing information to be cross-referenced for a tailor-made study on any subject that would not be explained in the initial classification plan.
The Report module of Proxem Studio allows you to create your tables and data visualizations independently.
Thanks to the many fully configurable widgets, you can set up as many dashboards as you want and distribute them within your company with ease. All visualizations are also exportable for integration into your presentations, and you can generate PDF reports ready to share or receive by email.
All the data produced in Proxem Studio can be exported, either via the Excel or CSV export functionalities available in the Explore and Report modules, or via the Rest APIs for perfect integration with your information system.
Proxem Studio simultaneously exploits artificial intelligence by learning (ML) and expert systems thanks to its powerful semantic rules engine.
Proxem Studio thus integrates machine learning algorithms at different levels for automatic language processing, in particular machine learning for language models based on neural networks and distributional analysis models, unsupervised learning for the discovery of the themes addressed in a corpus (word-embedding and clustering) and supervised learning algorithms.
Yes! For your projects, a project manager will accompany you throughout your usage of Proxem Studio.
A customer success manager will also offer workshops and training regularly to support you in the deployment in your organization, the implementation of good practices and taking into account your software development needs.
The price of the subscription to Proxem Studio depends on different parameters:
The overall volume of data to be processed per month / year
The number of users of the platform according to the profiles: Power Users access all the functionalities, in particular the modules making it possible to configure the semantic analysis; Analysts access the Explore and Report modules allowing them to conduct data analysis, create and disseminate dashboards etc.; Readers access only the Report module to consult the dashboards