NFT value evaluation method based on large model quantization and machine learning

Abstract

The invention relates to an NFT value evaluation method based on large model quantification and machine learning, and the method comprises the steps: building a mapping relation of NFT value and features, collecting historical data, carrying out the quantification of a part of features based on a large model, and training a machine learning model based on machine learning; and evaluating an NFT value based on the trained machine learning model. According to the method, by means of wide basic common knowledge of a large model, the ability of the method in the professional knowledge field is remarkably improved through fine adjustment, knowledge base retrieval, thinking chain and other technologies; a unified and credible method is constructed to objectively evaluate the value of the NFT, which is helpful for reducing occurrence of frying, easy for troubleshooting abnormal transaction behaviors, helpful for both transaction parties to achieve a value consensus, promoting the regularity of casting, circulation and collection of the NFT, and helping the healthy development of the NFT market.

Description

Technical Field

[0001] The present invention relates to the technical field of calculation, extrapolation or counting, and in particular to an NFT value assessment method based on large model quantification and machine learning. Background Art

[0002] NFT, the full name of which is Non-Fungible Token, refers to non-fungible tokens. In essence, it is a unique and reliable digital equity certificate in the blockchain network. It is a data object with multi-dimensional and complex attributes that can be recorded and processed on the blockchain. Each NFT represents an object in the virtual or real world (the same object can be represented by different NFTs). NFT is unique and tradable.

[0003] The objects represented by NFT are extremely diverse. For virtual items, they can be images, audio, video, game props, game characters, etc., or any objects in reality. Although NFT's representation of objects does not necessarily mean ownership, there are also cases where the ownership of NFT is used to indicate ownership. For example, in the game CryptoKitties (www.cryptokitties.co), holding an NFT means holding the game character corresponding to the NFT. In reality, there is also the idea of ​​using the ownership of NFT to indicate ownership. For example, the Chinese patent with publication number CN113537965A discloses a physical modeling and non-homogeneous virtual and anti-counterfeiting integrated system, which uses encryption calculations and non-fungible tokens (NonFungibleToken, NFT) methods to sell viewing rights or ownership rights through non-fungible trading platforms, allowing buyers to use it as a digital collection.

[0004] With the hot development of NFT in recent years, a large number of NFT casting, issuance and trading methods have been proposed, but there is a lack of a suitable NFT value assessment method in the current market. As an important carrier of the digital economy and an important part of the metaverse, a set of NFT value recognition system is indispensable.

[0005] A Chinese patent with publication number CN114283005A discloses a patent NFTization system and trading method, including a minting NFT subsystem, a patent evaluation subsystem, a patent NFT transaction price subsystem, a virtual land patent NFT display subsystem, etc., through the steps of minting patent NFTs, patent evaluation, and patent NFT transactions; the present invention realizes patent NFT value evaluation, and users who participate in the patent NFT behavior such as sharing, liking, and collecting will obtain the patent NFT income. The points mechanism encourages users to actively NFT the obtained patent certificates, and realizes the closed-loop system of filing on the NFT system and offline certificate delivery, giving the patent NFT value and the NFT quantity a technical solution as the source of virtual land value, thereby improving the efficiency of patent conversion; its method for patent NFT value evaluation is relatively rough and not universal.

[0006] Wang Juanjuan, Chen Hao, et al. published "Analysis of Factors Affecting the Value of NFT Digital Collections and Evaluation Framework - Based on FM Matrix Decomposition Algorithm" (Wang Juanjuan, Chen Hao. Analysis of Factors Affecting the Value of NFT Digital Collections and Evaluation Framework - Based on FM Matrix Decomposition Algorithm [J]. China Asset Appraisal, 2022 (7): 18-28. DOI: 10.3969 / j.issn.1007-0265.2022.07.004.) fully considered multiple dimensions and gave a value evaluation framework, but lacked a specific implementation plan. D. Costa et al.'s "Multimodal representation learning for NFT selling price prediction" (DOI: https: / / doi.org / 10.1145 / 3543507.3583520) constructed an evaluation model called the Merlin framework, which used images alone to evaluate NFTs and achieved a very high success rate. Kapoor A et al.'s "TweetBoost: Influence of Social Media on NFT Valuation" (DOI: 10.1145 / 3487553.3524642) used Twitter's social media data, and the machine learning model combined with social media data increased the accuracy by about 6% compared to not using it. However, this solution only uses quantitative social media data, does not consider the content factors of social media in the value assessment, and also ignores the object represented by the NFT itself. Summary of the invention

[0007] The present invention solves the problems existing in the prior art and provides an NFT value assessment method based on large model quantification and machine learning.

[0008] The technical solution adopted by the present invention is an NFT value assessment method based on big model quantification and machine learning. The method constructs a mapping relationship between associated NFT value and features, collects historical data, quantifies some features based on a big model, and then trains a machine learning model based on machine learning; and assesses the NFT value based on the trained machine learning model.

[0009] Preferably, the features include exchange rate and virtual currency value E, object artistry A, value RV of representative rights, popularity H, and historical price HP.

[0010] Preferably, some features are quantified based on the large model as follows:

[0011] The multimodal big model is used to quantify the artistry A of NFT’s metadata, the big model is used to quantify the value RV of the rights represented by NFT, and the big model is used to quantify the sentiment of NFT-related social media content to obtain the content coefficient and calculate the popularity H in combination with other quantitative data;

[0012] The numerical features of the object's artistry A, value RV representing rights, and heat H are obtained.

[0013] Preferably, the scoring of the artistry A of the object includes multiple scoring and committee scoring;

[0014] Multiple scoring means using the same prompt to ask the large model n times, and taking the average score as the final result;

[0015] The committee scores by using n large models fine-tuned based on different bases, removing the highest and lowest scores, and averaging the final scores;

[0016] The large model is fine-tuned using a dataset of artworks rated by experts, with mean square error (MSE) used as the loss function.

[0017] Preferably, it is determined whether holding an NFT represents having certain rights. If not, RV = 0. If so, obtain the reference value of the large model and calculate the RV based on the reference value. Retrieve the asset valuation knowledge base based on RAG technology to obtain relevant information on the value RV. Based on the thinking chain, obtain the value RV in multiple dimensions.

[0018] Preferably,

[0019]

[0020] Among them, CC i is the content coefficient of the i-th content, L i is the number of likes for the ith content, C i is the number of collections of the ith content, S i is the forwarding volume of the i-th content, Vi is the number of views of the ith content.

[0021] Preferably, the content coefficient is normally distributed, and a double threshold is defined to divide the value of the content coefficient into 1, 0 and -1 to represent positive, neutral and negative.

[0022] Preferably, the machine learning model is trained based on the exchange rate and virtual currency value E obtained from the relevant platform, the historical price HP of the current NFT, and the exchange rate and virtual currency value E, the artistry A of the object, the value RV of the representative right, the popularity H, the historical price HP, the NFT text information, and the NFT image as the data set to train the machine learning model;

[0023] The NFT value is derived from the quantitative data obtained using the trained machine learning model.

[0024] Preferably, the machine learning model includes an input layer and a fully connected layer.

[0025] Preferably, the machine learning model obtains a text feature vector based on the NFT text information, obtains a picture feature vector based on the NFT image, obtains a time series feature vector based on the exchange rate and the virtual value of the currency E and the historical price HP of the current NFT, and the text feature vector, picture feature vector, time series feature vector and numerical feature are collectively used as the input of the input layer.

[0026] The present invention relates to an NFT value assessment method based on large-model quantification and machine learning. The method constructs a mapping relationship between associated NFT values ​​and features, collects historical data, quantifies some features based on a large model, and then trains a machine learning model based on machine learning; and assesses the NFT value based on the trained machine learning model.

[0027] The beneficial effect of the present invention is that, with the help of the extensive basic common sense of the big model, its ability in the field of professional knowledge is significantly improved through fine-tuning, knowledge base retrieval, and thinking chain technologies; a unified and reliable method is constructed to objectively evaluate the value of NFTs, which helps to reduce the occurrence of speculation, facilitates the investigation of abnormal trading behaviors, helps the two parties to the transaction to reach a consensus on value, promotes the regularity of NFT casting, circulation, and collection, and helps the healthy development of the NFT market. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the present invention;

[0029] Figure 2 A schematic block diagram of the method of the present invention;

[0030] Figure 3 Schematic diagram of using machine learning model to evaluate the value of NFT in the present invention. DETAILED DESCRIPTION

[0031] The present invention is further described in detail below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto.

[0032] The present invention relates to an NFT value assessment method based on large-model quantification and machine learning. The method constructs a mapping relationship between associated NFT values ​​and features, collects historical data, quantifies some features based on a large model, and then trains a machine learning model based on machine learning; and assesses the NFT value based on the trained machine learning model.

[0033] The characteristics include the exchange rate and virtual currency value E, the artistry of the object A, the value of the representative rights RV, the popularity H, and the historical price HP.

[0034] In practical applications, functions can also be constructed based on features to satisfy V NFT =F(E,A,RV,H,HP), where F represents the functional relationship;

[0035] However, the function is difficult to fit in this process, so the features are divided into the part from which numerical features can be extracted and the part from which numerical features cannot be extracted.

[0036] Based on the large model, some features are quantified as follows:

[0037] The multimodal big model is used to quantify the artistry A of NFT’s metadata, the big model is used to quantify the value RV of the rights represented by NFT, and the big model is used to quantify the sentiment of NFT-related social media content to obtain the content coefficient and calculate the popularity H in combination with other quantitative data;

[0038] The numerical features of the object's artistry A, value RV representing rights, and heat H are obtained.

[0039] In the present invention, the influencing factors of artistry A, rights and rights value RV, and popularity H are difficult to quantify, so a large model is used to evaluate these three aspects and quantify them into scores; specifically:

[0040] (1) Use a multimodal big model to quantify the artistry of NFT’s metadata (representative object);

[0041] (2) The value of the rights represented by the NFT (if any) is quantified using a big model;

[0042] (3) Use a large model to quantify the sentiment of NFT-related social media content to obtain the content coefficient and calculate the heat H in combination with other quantitative data.

[0043] For (1), the scoring of the object's artistry A includes multiple scoring and committee scoring;

[0044] Multiple scoring means using the same prompt to ask the large model n times, and taking the average score as the final result;

[0045] The committee scores by using n large models fine-tuned based on different bases, removing the highest and lowest scores, and averaging the final scores;

[0046] The large model is fine-tuned using a dataset of artworks rated by experts, with mean square error (MSE) used as the loss function.

[0047] In this invention, the powerful multivariate information processing capability of the large model is used to process various images, texts, videos, audios or a mixture of different data types, and an artistic score of 0-100 is given. In order to ensure the stability of the score, multiple scoring and committee scoring methods are adopted;

[0048] Specifically, in order to ensure the stability of the artistic score A' of a single large model, multiple scoring is to use the same prompt to ask the large model n times, and take the average score as the final result, that is,

[0049]

[0050] The committee scores by using n large models fine-tuned based on different bases, removing the highest and lowest scores and averaging them to get the final score.

[0051]

[0052] Finally, we used the artwork dataset scored by experts to fine-tune the large model, using the mean square error (MSE) as the loss function.

[0053]

[0054] n is the number of samples, y + Give the expert score for the i-th sample, Score the large model for the i-th sample; adjust the parameters of the large model according to the value calculated by the loss function.

[0055] For (2), determine whether holding NFT means owning certain rights. If not, RV = 0. If so, obtain the reference value of the large model and calculate RV based on the reference value. Retrieve the asset valuation knowledge base based on RAG technology to obtain relevant information on value RV. Based on the thinking chain, obtain value RV in multiple dimensions.

[0056] In the present invention, we first consider whether holding NFT means owning certain rights. If yes, we use the retrieval and analysis capabilities of the big model to require the big model to give a value RV as a reference, and then use RAG technology to retrieve the asset valuation knowledge base to give a reference value RV to enhance the accuracy of the valuation. It should be noted that because the rights represented by NFT are extremely diverse, which can be stock ownership, antique ownership, copyright, etc., it is impossible to propose a general calculation method for RV. Therefore, we use the big model data and calculation mode to give a reference value for training and utilization.

[0057] Furthermore, the present invention uses the chain of thought (COT) method to improve the logic of the process in which the big model gives the value RV, requiring the big model to think in a specific chain of thoughts. For example, "Please evaluate the value based on [relevant information of the value RV], requiring consideration of multiple dimensions such as society, technology, ecology, and humanities, and give a value denominated in XX currency."

[0058] For (3),

[0059]

[0060] Among them, CC i is the content coefficient of the i-th content, L i is the number of likes for the ith content, C i is the number of collections of the ith content, S i is the forwarding volume of the i-th content, V i is the number of views of the ith content.

[0061] The content coefficient is normally distributed, and a double threshold is drawn to divide the value of the content coefficient into 1, 0 and -1 to represent positive, neutral and negative.

[0062] In the present invention, the popularity H is calculated using the public data of major social media platforms, and the data related to the predicted NFT is retrieved through the API opened by the social media platform. The data includes quantitative indicators such as likes L, reposts S, favorites C, and views V, and also includes text, images, videos and other content. The external sentiment dictionary is used as a knowledge base to improve the behavioral consistency and interpretability of the large model. The large model is used to first screen out the content related to the valuation of NFT, and then the large model is required to score the content sentiment factors for each social media message; the scoring rule of the large model is that the more positive the content sentiment, the closer the content coefficient CC is to 1, the more neutral the content sentiment, the closer the content coefficient CC is to 0, and the more negative the content sentiment, the closer the content coefficient CC is to -1, and it is proposed "Please evaluate [social media content], give the content coefficient, and only give a score."

[0063] In the present invention, if we only focus on quantitative indicators, we will lack understanding of the content, resulting in a completely opposite effect. Therefore, a formula for heat H is proposed, in which "+1" is processed to avoid the situation where a certain indicator is equal to 0.

[0064] In addition to the above three points, machine learning training machine learning models are also based on:

[0065] (4) The exchange rate and virtual currency value E obtained from the relevant platform,

[0066] (5) The current historical price of NFT, HP,

[0067] For (4), the data is obtained from the relevant institutions, which are the real-time exchange rates of various currencies against XX currency and the prices of major virtual currencies against XX currency at the time of evaluation; data E should be an ordered vector in which various currencies and virtual currencies are arranged in a fixed order.

[0068] For (5), it is a vector composed of all transaction records arranged in chronological order; since the currencies used in NFT transactions may be different and the value of virtual currencies fluctuates greatly, it should be uniformly converted to evaluate the value of XX coins at that time, that is,

[0069] HP=E : *P :

[0070] In the formula, E : is the exchange rate at that time or the exchange rate of virtual currency XX, P : The transaction price at that time.

[0071] In the present invention, the exchange rate obtained from the relevant platform is aligned with the virtual value of the currency E and the historical price HP of the current NFT in time, generally based on LSTM layer processing to capture the dynamic features in the time series and obtain the corresponding time series feature vector.

[0072] The obtained exchange rate and virtual currency value E, the object's artistry A, the value of the rights represented RV, the popularity H, the historical price HP, the NFT text information, and the NFT image are used as data sets to train the machine learning model;

[0073] The NFT value is derived from the quantitative data obtained using the trained machine learning model.

[0074] The machine learning model includes an input layer and a fully connected layer.

[0075] The machine learning model obtains a text feature vector based on the NFT text information, obtains a picture feature vector based on the NFT image, obtains a time series feature vector based on the exchange rate and the virtual value of the currency E and the historical price HP of the current NFT, and the text feature vector, picture feature vector, time series feature vector and numerical feature are collectively used as the input of the input layer.

[0076] In the present invention, a pre-trained convolutional neural network, such as ResNet50, is used to extract the image feature vector of the NFT image; a convolutional neural network is constructed, and the NFT image (the first frame of the video is captured) is used for classification, the classification category is the NFT series (Collection), and the input uses an RGB three-channel image. At the same time, the image is compressed and the center is cropped to 512*512. The NFT series has an extremely important influence on the image content of the NFT. In addition, the NFT collection prices in the same NFT series are mostly within the same range; after training to convergence, the classifier is removed, and the vector output by the last layer is used as the feature vector of the image.

[0077] In the present invention, a pre-trained BERT model is used to extract text feature vectors, and the text description information of the NFT text information is segmented and encoded.

[0078] In the present invention, the existing data are collected into a data set to train the machine learning model, and the fully connected layer is used to learn the mapping of the fused features, and finally the predicted continuous value is output; generally speaking, classification task training is performed first, that is, the current input is used to predict the range of the NFT value. After convergence, the model's classifier is replaced with a linear layer for regression task training. During evaluation, the obtained quantitative data is put into the machine learning model to obtain the predicted NFT value.

[0079] It should be noted that during the training and evaluation process, all features (vectors) have no order.

[0080] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0085] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A NFT value assessment method based on large model quantification and machine learning, characterized by: The method constructs a mapping relationship between the associated NFT value and features, collects historical data, quantifies some features based on a large model, and then trains a machine learning model based on machine learning; and evaluates the NFT value based on the trained machine learning model.

2. According to claim 1, a NFT value assessment method based on large model quantification and machine learning is characterized in that: The characteristics include the exchange rate and virtual currency value E, the artistry of the object A, the value of the representative rights RV, the popularity H, and the historical price HP.

3. According to claim 2, a method for evaluating the value of NFT based on large model quantification and machine learning, characterized in that: Based on the big model, some features are quantified as follows: the artistry A of NFT’s metadata is quantified by the multimodal big model, the value RV of the rights represented by NFT is quantified by the big model, and the sentiment of NFT-related social media content is quantified by the big model to obtain the content coefficient and calculate the popularity H in combination with other quantitative data; The numerical features of the object's artistry A, value RV representing rights, and heat H are obtained.

4. According to claim 2, a NFT value assessment method based on large model quantification and machine learning is characterized in that: The scoring of the artistry A of the object includes multiple scoring and committee scoring; Multiple scoring means using the same prompt to ask the large model n times, and taking the average score as the final result; The committee scores by using n large models fine-tuned based on different bases, removing the highest and lowest scores, and averaging the final scores; The large model is fine-tuned using a dataset of artworks rated by experts, with mean square error (MSE) used as the loss function.

5. According to claim 2, a NFT value assessment method based on large model quantification and machine learning is characterized in that: Determine whether holding NFT means owning certain rights. If not, RV = 0. If so, obtain the reference value of the large model and calculate RV based on the reference value. Retrieve the asset valuation knowledge base based on RAG technology to obtain relevant information on value RV. Based on the thinking chain, obtain value RV in multiple dimensions.

6. According to claim 2, a NFT value assessment method based on large model quantification and machine learning is characterized in that: Among them, CC i is the content coefficient of the i-th content, L i is the number of likes for the ith content, C i is the number of collections of the ith content, S i is the forwarding volume of the i-th content, V i is the number of views of the ith content.

7. According to claim 6, a method for evaluating the value of NFT based on large model quantification and machine learning, characterized in that: The content coefficient is normally distributed, and a double threshold is drawn to divide the value of the content coefficient into 1, 0 and -1 to represent positive, neutral and negative.

8. According to claim 3, a NFT value assessment method based on large model quantification and machine learning is characterized in that: Based on machine learning training, the machine learning model is also based on the exchange rate and virtual currency value E obtained from the relevant platform, the historical price HP of the current NFT, and the exchange rate and virtual currency value E, the artistry A of the object, the value RV of the representative rights, the popularity H, the historical price HP, the NFT text information, and the NFT image as the data set to train the machine learning model; The NFT value is derived from the quantitative data obtained using the trained machine learning model.

9. The NFT value assessment method based on large model quantification and machine learning according to claim 8 is characterized by: The machine learning model includes an input layer and a fully connected layer.

10. The NFT value assessment method based on large model quantification and machine learning according to claim 9 is characterized in that: The machine learning model obtains a text feature vector based on the NFT text information, obtains a picture feature vector based on the NFT image, obtains a time series feature vector based on the exchange rate and the virtual value of the currency E and the historical price HP of the current NFT, and the text feature vector, picture feature vector, time series feature vector and numerical feature are collectively used as the input of the input layer.

Images