BrandGPT (GPT Neural Network)

The main articles are:

• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2024

Availability of a smart AI assistant

On April 16, 2024, Brand Analytics announced that an AI analyst assistant, BrandGPT, had become available to users. The solution is based on a large language generative model designed to quickly analyze social media and media in the interests of business. The first working prototype BrandGPT was presented to the public in the fall of 2023. And in January 2024, the Federal Intellectual Property Service officially registered BrandGPT software. Before the public launch, the AI assistant underwent closed testing in 30 companies. BrandGPT is designed to solve the problem of significant time and human costs of analyzing really big social media data. Regardless of the amount of data analyzed, the BrandGPT service generates analytical answers for typical tasks in the field of marketing research in a matter of minutes and allows you to quickly make management decisions.

𝓲

Brand Analytics

{{quote 'author
= Natalia Sokolova, Managing Partner of Brand Analytics' BrandGPT is addressed not only to direct users of the system, but also to a wider range of managers and specialists: product managers, brand managers, marketing directors, PR directors, etc. Everyone can ask BrandGPT a question and find out what is important about the company, person or product, as well as get hypotheses for deciding on the next steps. In our experience, for April 2024, the most popular requests to analysts are, just, "What is happening?" and "What are your hypotheses?." And answers are expected as quickly as possible.}}

According to the company, BrandGPT is the own development of the Brand Analytics team. Its peculiarity, according to the developers, lies in the relevance, security and marketing specialization of the AI assistant.

𝓲

Brand Analytics

The AI assistant is able to analyze not only the past, but also the present - the current information field around a brand, product or other object of analysis. At the same time, BrandGPT ensures the security of the processed data. In the process of analysis, data is not transferred to third-party services - all processing is carried out inside the Brand Analytics system infrastructure on its own servers in Russian data centers. When using BrandGPT, the risk of data leakage, which is possible, for example, when working with popular foreign neural networks, is excluded. Marketing specialization is another feature of BrandGPT. The model is trained on its own datacets from tens of millions of Brand Analytics messages and knows how to understand social media and media data. The neural network is familiar with marketing tasks and can answer the relevant question according to the data collected in Brand Analytics.

In addition, BrandGPT is a responsible language model that minimizes the tendency to hallucinations. Factchecking with multi-stage verification of the reliability of the answer in most situations blocks fictitious facts. And the AI assistant argues any conclusions using examples of reports from the media and social media, which in the overwhelming number of cases are indicative of analytics and help to quickly understand what is happening within the framework of the topic under study, Brand Analytics said.

One of the decisive factors for optimizing the language model is the availability of data for its training. Brand Analytics is an operator of social media data, which allows you to build the process of further training the model in the optimal way.

{{quote 'author
= told Grigory Ostrovsky, STO Brand Analytics' BrandGPT is not one language model, it is a complex, technologically complex solution. In the AI assistant, we have provided a certain protection against issues that have nothing to do with the topic of monitoring. We taught the model to adapt user requests so that the machine understands them better. Since the main task of our assistant is not content generation, but analytics, we continue to work on factchecking responses to minimize the hallucinations inherent in all GPT solutions.}}

Brand Analytics users can work with the BrandGPT service using pre-configured scripts to solve common problems, can ask the AI assistant recommended questions, or use questions in free form. When you select one of the ready-made scenarios, the data for the selected time period is filtered to obtain the desired data set, and the desired ML model is connected for analysis. In the most commonly used scenario, BrandGPT will automatically analyze hundreds of thousands or millions of messages in minutes and show a summary in the form of abstracts with supporting examples.

In addition to the response from BrandGPT, you can study the examples provided and formed for the analysis of datacets. Among the ready-made scenarios: basic news, topics of positive and negative, advantages and disadvantages according to users, etc. When you activate the script, you can refine the analysis object. The set of ready-made scenarios for effective work with the AI assistant will constantly expand. The second way to work with BrandGPT is to ask the recommended or your own question. The task of the AI assistant can be formulated in free form. For example, "tell us what our customers have written about the product over the past month." Most efficiently, BrandGPT is able to analyze monobrand topics, as well as topics related to specific companies, persons, products or services, the company specified.

BrandGPT is already available to users of the Brand Analytics system, starting with the SP5 tariff. In the near future, it is planned to connect an AI assistant for younger tariffs. In the future, it will be possible to buy request packages in addition to packages available in tariffs.

In the future, it is planned to expand the set of ready-made scenarios. Add an analysis of the mention peaks, for example, in response to the question: "What are the bursts of negativity associated with?." Implement filtering and sorting management through queries: "Show the most audience messages." Teach the assistant to take into account message metrics, for example, in the answer to the question: "What do they write in the most cited publications?" And the developers are also working to teach BrandGPT how to use statistics.

Registration in Rospatent

Brand Analytics has officially registered BrandGPT with the Federal Intellectual Property Service. The developer announced this on January 24, 2024.

Strengthening technological sovereignty was announced as one of Russia's key priorities in 2023. An important component of such sovereignty is leadership in the field of AI.

BrandGPT's official release and press conference are scheduled for March 2024. For January 2024, user pre-release testing is taking place.