The majority of suppliers of conversation systems provide solutions dedicated to one communication channel (e.g. smartphones, internet messengers or chat). In the times of multi-channel communication (omni-channel) it carries the necessity to maintain and integrate various solutions, and the analytics is far from perfect.
When creating the Omni-Chatbot we built a solution, which based on a one-time prepared knowledge database is able to work in various communication channels and provide a full image of the interactions with customers in one place. Such an approach is cheaper to install (one software installation, one integration works), impalement (one-time configured database, one training for the administrators) and maintain (one database in one place, one analytical tool collecting data from various channels).
Enterprise class architecture
OCP consists of many elements adjusted to our customers’ needs and architecture, in which it is to be implemented. The core of the solution is an application server based on microservices and the Administrative panel server. Depending on the selected interface automatic speech recognition (ASR), text-to-speech (TTS), interactive voice response (IVR), or middleware (for instant messengers) modules are added.
The data collected by OCP (log files) can be subjected to additional analytics by external solutions, such as Kibana or IBM Watson Explorer. Additionally, we also make the application programming interface, which enables creation of own solutions and interface (e.g. mobile applications or robots), available to the customers.
Advanced NLU tools
Twelve years of experience, realization of more than 130 projects and the ongoing development of the system contributed to the fact that our natural-language understanding algorithms (NLU) are one of the best in the world. Moreover, from the very beginning they were designed to handle the complexities of the Slavic languages grammar (i.e. flexible syntax grammar), rich inflection and word-formation diversity (synonyms, diminutives).
OCP understands users’ statements trough detecting intentions together with taking into consideration the context and information collected during conversation or derived from external systems. The conversation might be both of a ‘casual’ character, and also be a part of a structured dialog (e.g. authorization process). The way in which the system interprets a user’s statements can be controlled using the Administrative panel, which is equipped with tools facilitating the configuration process, such as intention patterns, ontologies, synonyms or word inflection databases.
Configuration of the solution might be entirely conducted by trained members of our customer’s team or by Stanusch Technologies specialists. Advanced Administrative panel makes it possible to configure knowledge database, establish business rules concerning how the solution works and integrate the solution with external data sources. The panel is also equipped with tools automatizing the knowledge base testing (manual tests, automatic tests, conflict detection, etc.) and tools facilitating system learning.
The analytical part supplies current information on the way the platform is working and categorizes all conversations, whereas the machine learning techniques are used to detect trends and generate managerial alerts if abnormal situations occur (e.g. sudden increase in the number of complaints).