Artificial emotional intelligence

Natural Language Processing (NLP)

Our 27 proprietary and patented NLP modules that are independently developed contain the industry's most advanced self-organizing tree algorithm (SOTA) and pre-training model. With the industry’s most complete industrial environment-oriented NLP model and corpus and based on language knowledge, we have independently developed a language model and a contextual understanding model according to the characteristics of dialogs (spoken language, short text) and long texts, respectively. Model compression has been performed on the basis of common models such as BERT, RoBERTa and XLNet, with the training and reasoning cost reduced by 50 times, solving the problem of difficulty in implementing a large-scale deep learning model.

Supporting the world's major languages, it has accumulated mature and rich NLP capabilities in simplified Chinese, English, Japanese and other languages, and can easily respond to the challenges of the global market.

The model can be optimized through continuous feedback and automatic accumulation. With a lot of training and optimization and linguists' experience in industry linguistics, we provide mature and ready-to-use industry models in six major industries to empower industry AI solutions through industry knowledge and language models.

Artificial emotional intelligence

Knowledge Engineering

Knowledge engineering has the ability to mine knowledge, build knowledge graph and ontology based on unsupervised learning. By mining and analyzing large-scale data and applying accumulated industry knowledge and NLP ability, the traditional human-based knowledge engineering process can be automated, which greatly improves the efficiency and lowers the threshold of knowledge graph creation.

It can analyze and process massive heterogeneous data, parse, classify, predict, cluster and analyze structured and unstructured data, support intelligent search, intelligent recommendation, intelligent prediction and other applications, and solve business problems such as risk control, anti-fraud, anomaly discovery, prediction and alarm, and root cause analysis.

Intelligent business process automation is realized through AI + RPA, and complex and tedious business processes are handed over to the machine. With the process editor, you can quickly build a business process automation task, which can reference thousands of mature industry AI models and greatly reduce the development threshold.

Artificial emotional intelligence

Deep Learning

The neural network architecture and parameters are automatically optimized by automatic neural network architecture search (NAS), so as to realize automatic model selection and model optimization without manual parameter optimization. Based on multi-model automatic ensemble and super-parameter optimization ability, we can find the best model combination, reduce the structural risk and maximize the utility of data.

Machine learning platform combines model management, AutoML capability and distributed computing capability of heterogeneous computing platform to provide comprehensive model training and reasoning services. It is a perfect data operation platform with automatic expansion, automatic data conflict detection, data quality inspection, tagging, data optimization and other capabilities.

Through reinforcement learning and unsupervised learning, the system can realize large-scale data mining and automatic strategy learning based on business rules. It can automatically discovered knowledge and find the optimal business strategy through massive strategy simulation and knowledge refinement from massive data.

Artificial emotional intelligence

Intelligent Speech Technology

The fully self-developed ASR / TTS model, combined with self-collected training data, SOTA speech recognition algorithm and the industry's best language model with NLU capability, optimizes the model performance for different business scenarios. It has a complete training platform, which can support complete custom training for vertical fields.

Support full-duplex media control, and support voice capabilities such as silence and interruption required for two-way communication.

Accurately judge the emotion of users based on voice emotion and text emotion recognition.

Artificial emotional intelligence

Multi-modal Affective Computing

Face micro-expression, emotion recognition, emotion attendance, body movement recognition

Speech emotion and voiceprint recognition

Emotion recognition based on facial expression, voice and semantics, accurately judging emotional state

Artificial emotional intelligence

Text Data Middleground

Text comparison, text parsing, text duplicate checking, text error correction, text correction, intelligent writing

With deep learning and NLP ability, it can accurately understand the user's intent and match the content. Unstructured text is represented in vector space and combined with big data to produce retrieval ranking that can understand semantics.

Based on the user profiles created through heterogeneous data refinement and mining and reasoning capabilities of knowledge graphs, it can identify and reason about complex user business scenarios, accurately determine the actual needs of users to achieve accurate recommendations.