Interpretation of and Advice on the Draft Revision to the Patent Examination Guidelines (2021.8) (III)
The Draft Revision to the Patent Examination Guidelines (Draft for Comment) was released to seek public comments as of August 3 this year. The contents regarding application for invention patent on computer programs in Chapter 9, Part II are partly revised, with several modified or newly-introduced examination samples as reference examples.
The Revision specifies the examination criteria for patents in the computer programs related field, for which the writer will make interpretations and provide comments in terms of operational practice in patent prosecution.
I. Part II, Chapter 9, Section 5.2: “Drafting of Claims”
It is added to the Draft Revision that “computer program products shall be construed as software products providing solutions mainly through computer programs” and that “computer program products” may be the objects of protection of claims.
As a consequence of this revision, the claims of invention patent applications concerning computer programs can be drafted in the form of four groups as follows in the future:
Apparatus, equipment and system claims
Computer program product claims
Storage medium claims
As technology evolves, this Revision keeps effective to meet diversified demands of innovation subjects from all walks of life, through adding “computer program product claims”. Criteria for drafting and examining invention patent applications involving computer programs will be further in line with international standards.
Second, the infringing product of the invention concerning computer programs is specific and the right owner only needs to take actions against the manufacturer of software rather than hardware such as processors and memories.
Moreover, with the rapid development of cloud computing, most computer programs or software are stored in the cloud. Since computer program product claims may allow the patentee to obtain relevant computer program product from the Internet, rather than by controlling the infringer’s devices such as computers or servers installed with the computer software, it will facilitate the evidence collection of the patentee to prove the infringement.
Last, patent applicants and agents could add contents of computer software products to specifications during the transition period before the finalization of the Draft Revision to the Patent Examination Guidelines, and add the computer program product claims into the patent claims in later amendment on their own initiative.
II. Part II, Chapter 9, Section 6.1.2.: “Examination under Article 2.2 of the Patent Law”
Examination criteria for technical solutions provided in Article 2.2 of the Patent Law have been specified by the following four examination samples in section 6.2 of the revised Guidelines.
A method for training Deep Neural Network (DNN) model
This invention patent application provides a method for training Deep Neural Network model, wherein the least time-consuming training program is selected from multiple training programs for a fixed size of training data to solve the problem of slow training speed resulting from the constant application of the same single-processor or multi-processor training program to the training data of all sizes.
A method for training Deep Neural Network model, wherein:
the training time of the changed training data in the preset candidate training programs is calculated respectively at the change of the training data size;
the least time-consuming training program is selected from the preset candidate training programs as the optimal training program for the said changed training data, wherein the said candidate training programs include the single-processor training programs and data-parallel-based multi-processor training programs;
the said changed training data are input in the said optimal training program for model training.
Analysis and conclusion
The solution relates to a method for training Deep Neural Network (DNN) model, wherein single-processor or multi-processor training programs with diversified data processing efficiency are selected for training data of different sizes in order to solve the problem of slow training speed. The model training method with specific technical connection to the internal structure of computer system promotes the performance of hardware in the course of model training, thus achieving a natural technical effect of improving the internal performance of computer system. Therefore, the solution in the invention patent application belongs to technical solutions as provided for in Article 2.2 of the Patent Law and is the subject matter of patent protection.
A method for analyzing the tendency to use electronic coupons
It is a strategy of merchants to solicit customers through distributing electronic coupons (“e-coupons”) of various kinds. However, such aimless distribution of e-coupons not only fails to reach those in actual need, but also imposes burden on customers to browse and filter the coupons. The invention patent application provides a method for building a model to identify the tendency to use e-coupons, which can precisely build an e-coupon use tendency identification model through analyzing the types of e-coupons and customer behavior, so as to determine customers’ tendency to use e-coupons more accurately and make customer-oriented distribution of e-coupons to improve utilization of e-coupons.
A method for analyzing the tendency to use e-coupons, characterized in that it includes the following steps:
categorizing e-coupons into different types based on the information on them;
obtaining sample data of customers based on usage scenarios of the e-coupons;
extracting customer behavior features from the said sample data of customers according to customer behavior, including web page browsing, keywords searching, favorite adding, adding items into the cart, e-coupon purchasing and using; training the e-coupon use tendency identification model regarding different types of e-coupons using sample data of customers as training samples and customer behavior features as property labels;
predicting the probability of use of e-coupons through the trained e-coupon use tendency identification model to get customers’ tendency to use different e-coupons.
Analysis and conclusion
This solution involves a method for building a model to identify the tendency to use e-coupons, which discloses the internal correlation between customers’ behavior and their tendency to use different e-coupons through processing relevant big data. By categorizing electronic coupons, collecting sample data, identifying behavior features and conducting model training, the internal correlation between customer behavior features and the tendency to use e-coupons is dug out, namely such behavior features as long-time browsing, repetitive searching and frequent use of e-coupons reflect a higher tendency to use the corresponding types of e-coupons. The technical problem of improving the accuracy of analysis of customers’ tendency to use e-coupons is therefore solved based on such internal correlation complying with the laws of nature and the relevant technical effect is achieved. Therefore, the invention application belongs to technical solutions as provided for in Article 2.2 of the Patent Law and is the subject matter of patent protection.
A knowledge graph inference method
Knowledge graph plays an important role in many NLP applications such as question answering system and semantic search. Due to the uncertainty of knowledge acquisition, knowledge graph built based on entity recognition and relation extraction technologies may be incomplete. If any errors are contained in the knowledge graph, the application will return erroneous responses. This invention patent application provides a knowledge graph inference method based on relation attention.
A knowledge graph inference method based on relation attention, comprising:
obtaining initial embedding representations of nodes in knowledge graph and projecting them into higher dimensions to acquire high-dimensional embedding representations, wherein the said nodes are entities in the knowledge graph constructed through entity recognition and relation extraction of knowledge. The said knowledge specifically refers to interrelated knowledge in question answering system and semantic search, the said entities here are textual data obtained from natural language texts using naming entity recognition tools and the said initial embedding representations are vectors obtained from the said textual data by a word embedding model;
obtaining a collection of neighbor nodes of the target nodes in the said knowledge graph, and constructing neighbor subgraphs according to the type of relation between the said target nodes and the neighbor nodes in the said collection of neighbor nodes;
obtaining the neighbor embedding representations of the said target nodes embedded in the neighbor subgraphs according to the high-dimensional embedding representations of the said target nodes and the neighbor nodes in the said neighbor subgraphs;
aggregating the high-dimensional embedding representations of the said target nodes and the said neighbor embedding representations to obtain the aggregated embedding representations of the target nodes;
fusing the said aggregated embedding representations based on the first attention score of each said neighbor subgraph to obtain the fusion embedding representations of the said target nodes;
Calculating scores of the triples corresponding to the said target nodes according to the said fusion embedding representations, and making triple inferences based on the scores.
Analysis and conclusion
This solution relates to a knowledge graph inference method based on relation attention, whereinsuch technical data as the textual data of natural language and the semantic information are processed in each step and a knowledge graph is constructed to make knowledge graph inferences through entity recognition and relation extraction of interrelated knowledge in question answering system and semantic search. This solution addresses technical issues of enriching semantic information and increasing accuracy of inferences during text embedding and semantic search, using a natural technical means to achieve a corresponding technical effect. Therefore, the solution in the invention patent application belongs to technical solutions as provided for in Article 2.2 of the Patent Law and is the subject matter of patent protection.
A method for forecasting the price of financial products
In most existing methods for forecasting the price of financial products as advised by experts based on their experience, the accuracy and time efficiency of the prediction are not high. The invention patent application provides a method for forecasting the price of financial products, which can train the neural network model with historical price data of financial products so as to predict future price trend of the financial products.
A method for forecasting the price of financial products, characterized in that it includes the following steps:
training the neural network model to obtain a price forecasting model with N+1 daily indicator historical price data of financial products, wherein the N daily indicator historical price data are sample input data and the last daily indicator is the sample result data;
predicting the price of financial products of the next day based on the said price forecasting model and the recent N daily indicator historical price data.
Analysis and conclusion
The solution relates to a method for forecasting prices of financial products, which processes the financial- product-related big data and digs out the internal correlation between price data of the financial products within a recent period and those in the future with a neural network model. However, such internal correlation in compliance with the laws of nature does not exist, since the price trend of financial products is subject to economic rules rather than being determined by its previous performance. Since the problem to be solved by this solution is financial product price forecasting, which is not a technical problem, the relevant effect obtained therefrom cannot be considered as a technical effect as well. Therefore, the invention application does not belong to technical solutions as provided for in Article 2.2 of the Patent Law and is not the subject matter of patent protection.
First, the general principle to determine whether an invention involving computer programs complies with Article 2.2 of the Patent Law remains unchanged, which continues to depend on the ability of the overall solution to address any technical issues and obtain any technical effects.
Second, with respect to the innovation involving computing models and algorithms, claims of the patent application shall demonstrate technical effects such as improved computational speed or accuracy. According to previous examination rules, the specific application fields of applied-for patents involving AI network model shall be clarified, such as image or video processing, while as shown in newly-added examination samples, it is not necessary for solutions involving AI algorithms to apply to any specific application area or address any subject matter. Subsequently, the protection scope of claims would be extended without the limitations of application fields including image and video processing.
Moreover, for applications involving behavior analysis or financial issues which are liable to be defined as “rules and methods for intellectual activities”, it is necessary to reflect the technical improvements, rather than the mere realization or representation of certain rules with computer programs.
III. Part II, Chapter 9, Section 6.1.3.: “Examination on novelty and inventive step”
With reference to the two examination samples in Section 6.2, the Draft Revision elucidates criteria for examination on inventive step, i.e. the algorithmic features or business rules and methods which combine technical features or have functional support to or interaction with the technical features shall be taken into consideration.
A logistics delivery method
This invention patent application is to solve the problems of effectively improving the efficiency and reducing costs of goods delivery. Upon arriving at the delivery area, deliverers can simultaneously send push pick-up notifications through the server to the terminals of the ordering users in a specific delivery area, thus realizing the purposes of improving delivery efficiency and reducing costs of delivery.
A logistics delivery method to improve logistics delivery efficiency by sending bulk pick-up notifications to users, wherein:
where there is a need to notify users to pick up, the deliverer sends notifications of goods arrival to the server from a handheld logistics terminal;
the server sends bulk notifications to all ordering users in the delivery area of the deliverer;
the ordering users receiving notifications pick up their goods accordingly;
the server sends the bulk notifications in the following way: identifying all target orders’ information corresponding to the deliverer ID and within the delivery range centered in the current location of the said logistics terminal according to the deliverer ID, current location of the logistics terminal and the corresponding delivery scope sent by the logistics terminal first, then pushing notifications to the ordering users’ terminals corresponding to the ordering users’ accounts contained in all target orders.
Analysis and conclusion
Reference document 1 makes public a logistics delivery method, wherein the deliverer uses a logistics terminal to scan the barcode on the delivery sheet and send the scanned information to the server to notify the arrival of the goods; the server extracts the ordering user’s information from the scanned information and sends out the notification; the notified user picks up the parcel following the message. The technical solution described in the invention patent application is different from that of the reference document 1 in that to realize bulk notifications to users of the goods arrival, the data structure and data communication method among the server, the logistics terminals and the users’ terminals in the solution of the invention application are adjusted, and the pick- up notification rules and specific bulk notification realization method support and affect each other in function. Compared with the reference document 1, the technical problem to be solved by the invention patent application is determined to be how to notify the users of the goods arrival more efficiently so as to improve the efficiency of goods delivery. By this way, deliverers can operate in a more convenient way and the ordering users can receive the pick-up notices in a timelier manner, which improves the experience of both the delivery and the acceptance parties. Given that the solution in this patent application can (i) promote user experience through a combination of pick-up notification rules and specific bulk notification methods, as well as adjustments on data structure and data communication methods, which functionally support and affect each other, and (ii) improve delivery efficiency based on the increased notification efficiency, which can be regarded as a “technical effect”, it has beneficial effects which prior arts do not possess. Since prior arts have not provided the technical inspiration that the technical solution in the invention patent application can be obtained through improvements on the abovementioned reference document 1, it can be concluded that the claimed technical solution of the invention possesses inventive step.
A neural network parameters matching method
To meet the demands of diversified application scenarios, it is necessary to design different neural network structures and conduct a series of operations on a certain type of computational structure, desirably in an effective manner with relatively low hardware costs. This invention patent application provides a method for matching neural network parameters, which can obtain neural network parameters in the normative form and map the operations on neural network structures to those supported by the computational structure, so as to simplify the design and realization of neural-network-related hardware.
A neural network parameters matching method, comprising:
selecting multiple dimensions to project weighting parameters on each of neural network layers in at least one layer of the neural network;
determining sizes of the said weighting parameters in each of the said multiple dimensions;
determining a collection of candidate values of the target size of the said weighting parameters in each of the said multiple dimensions according to the usage of the hardware supporting neural network operations;
selecting a sub-collection of all candidate values no less than the sizes of parameters in corresponding dimensions, and confirming the minimum value in the said sub-collection of candidate values to be the target size on the corresponding dimension;
where the sizes of the said weighting parameters in at least one of the multiple dimensions are smaller than the target size(s) in the corresponding dimension(s), performing the padding operation to such weighting parameters in such dimension(s), so as to ensure that after padding, the sizes of the said weighting parameters in each dimension are equal to the target size in the corresponding dimension.
Analysis and conclusion
Reference document 1 makes public a method for designing neural network processor through finding unit library from the built neural network component library according to the topological structure of neural network, the weighting and dimensional parameters on each neural network layer and the parameters of hardware resource constraints, etc., accordingly generating the hardware description language (HDL) code corresponding to the neural network processor of the neural network model and converting the said HDL code into hardware circuit of the said neural network processor. In this solution, neural network feature data and weighting data are divided into proper data blocks for centralized storage and access. The solution proposed in this invention patent application differs from the solution proposed in the reference document 1 in the following steps: determining sizes of each layer’s weighting parameters in each dimension, determining a collection of candidate values of the target size of weighting parameters in each dimension according to hardware usage, selecting a sub-collection of candidate values in the corresponding dimension and determining the minimum value to be the target size, and then performing padding operation to weighting parameters of which the sizes in at least one of the dimensions are smaller than the corresponding target sizes. According to application documents, this solution makes up the gap between the actual sizes of weighting parameters and corresponding target sizes, thus enabling the hardware compatible with neural network structures to process relevant data efficiently when performing neural network operations, which is realized by virtue of algorithms in the solution. Therefore, the said technical and algorithmic features for matching neural network parameters support and affect each other functionally. Compared with reference document 1, this invention patent application addresses the practical technical issue of ensuring the high efficiency of the hardware to perform neural network operations. The foresaid solution, which can improve the computing efficiency of hardware through matching neural network parameters, has not yet been disclosed by any other reference documents and does not belong to common sense in this field.Since in general prior arts have not provided the technical inspiration that the technical solution in the invention patent application can be obtained through improvements on the abovementioned reference document 1, it can be concluded that the claimed technical solution of the invention possesses inventive step.
The Draft Revision specifies the criteria for examination on inventive step, which shall take into account the correlation between technical features and non-technical features such as algorithms, business rules and methods, rather than simply denying the role of non-technical features.
Firstly, with respect to invention patent applications involving computer algorithms or business modes, applicants shall not only pay attention to combining technical means in the technical solution, but also emphasize the corresponding technical effects of this solution.
Secondly, with respect to inventions combining computers with computer algorithms or business modes, applicants may focus on the illustration of the combination between the algorithms and business rules and the specific technical parameters and structural components of computers, thus highlighting impacts of the algorithms and rules on computational speed (faster) and efficiency (higher accuracy/less amount).
Moreover, descriptions of the invention’s effect in specification with a sole emphasis on user experience, excluding technical effects, may run a risk of failing to meet the inventive step requirements of the examination, which is therefore not recommended.
In conclusion, with respect to Article 2.2 of the Patent Law and novelty and inventive step, the examination on computer program inventions becomes more reasonable. When drafting invention patent applications involving computer programs or business modes, applicants shall demonstrate the three elements of “technical means, problems and effects” to satisfy the provisions of Article 2.2 of the Patent Law and offer grounds for replying to office actions, particularly those involving inventive step.
The Watson & Band website is intended for informational purposes only. Nothing in this site is to be construed as creating an attorney-client relationship between the reader and Watson & Band or as offering legal advice on any specific matter. Since we are not providing legal advice through this website, you should not act upon any information that you might receive here without first seeking professional counsel. No client or other reader should act or refrain from acting on the basis of any information contained in the Watson & Band website without seeking appropriate legal or other professional advice based on the particular facts and circumstances at issue.