3719301 - Language Detection Challenges in Multi-Language Email Content for Ticket NLP Classification

In the Ticket NLP Classification scenario of SAP Cloud for Customer, inbound emails are processed using Machine Learning models that perform: Language detection

SAP Cloud for Customer

1. C4C receive customer emails written in English. However, some emails contain:
   • Name in different language
   • Addresses or location references in non-English languages
2. Even though the full email itself is English, the presence of such linguistic tokens sometimes triggers the language detector to incorrectly classify the message as : Spanish/Italian/English or any other language

According to the Machine Learning documentation for Ticket NLP Classification:

• The system uses statistical and neural language detection methods.

• Detection is sensitive to character patterns, name structures, and frequently occurring words associated with a language corpus.

• Short texts or text with named entities from multiple languages can mislead the classifier.

1. This is expected behavior.
2. The NLP model used in the system is designed to process only one supported language at a time within the ticket descriptionmultilingual content in a single text is not supported.
3. When the model detects mixed languages, it becomes unable to reliably determine the primary language for processing.
4. In such cases, the linguistic analysis is impacted, and the model may fail to correctly interpret the text as English.
5. Therefore, the presence of even a few foreign-language words can cause the model to misidentify or fail to process the language correctly

Ticket NLP Classification | SAP Help Portal

language detection, multi-language emails, ticket NLP classification, machine learning challenges, C4C language processing, mixed language content, neural language detection, statistical language detection, multilingual text issues, language misclassification, NLP model limitations, foreign language tokens, linguistic analysis, email processing errors, language corpus , KBA , LOD-CRM-ML , C4C Machine Learning , LOD-CRM-ML-SRV , Machine learning for service cloud , Problem