Classifying information

Abstract

In an embodiment, a computer-implemented method (100) of classifying invention-related information in relation to one or more sources is provided. The method comprises identifying (102) an indication of a Distinctive Causal Output, DCO, and one or more Essential Causal Features, ECFs, in invention-related information. The method further comprises classifying (104) the invention-related information according to whether or not the indication of the DCO and one or more ECFs indicates that the invention-related information belongs to either: a first class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are different to one or more ECFs of the one or more sources; or a second class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are the same as one or more ECFs of the one or more sources. The method further comprises providing (106) an indication of whether the invention-related information has been classified as belonging to the first class, the second class or does not belong to either class.

Description

[0001] CLASSIFYING INFORMATION

[0002] TECHNICAL FIELD OF THE INVENTION

[0003] The invention relates to computer-implemented methods, an apparatus, a computer program and a machine -readable medium for, particularly but not exclusively, classifying information.

[0004] BACKGROUND OF THE INVENTION

[0005] Prior art searches are an integral part of the patent system as a whole, as they enable establishing the ‘novelty’ and the ‘inventive step’ of a given invention by comparison to earlier disclosures / state of the art. Searches are carried out at different temporal points, including:

[0006] While preparing and prosecuting patent applications by patent attorneys and patent examiners.

[0007] While addressing post-grant questions regarding the validity or invalidity of a granted patent.

[0008] While addressing pre-use questions regarding whether a given product does or does not infringe on someone else’s patent rights.

[0009] These searches have always been guided by some system of classification. Historically, the searches were done manually, primarily by reviewing printed documents, before the process became increasingly computerized. In the first instance, this involved digitizing millions of patent documents, as to make them available to the searchers. As the computerization process progressed, (most) new patent applications now enter the system already digitized. Furthermore, increasingly sophisticated classification systems and software tools have been introduced to facilitate the process. Currently, millions of digitized patent documents are stored within patent databases, at institutions such as World Intellectual Property Organization (WIPO), United States Patent and Trademark Office (USPTO), European Patent Office (EPO), and United Kingdom Intellectual Property Office (UKIPO). Government-owned data bases are normally accessible free of charge, whilst commercial databases are normally accessible for a fee. These patent databases are often accessible online, and are searchable using the International Patent Classification [IPC], and in some cases ‘keyword’ searches. Undoubtedly, these technological and operational advances have facilitated innovation. However, problems persistently exist. The primary problem can be described as escaped relevant information. Despite the sophistication of the search systems and the hard efforts of those using them, crucially important relevant information escapes being captured in some searches. This is evident by a number of facts. For a start, different patent office searches often capture different prior art documents for the same invention. Sometimes, even the same patent office updates their searches by including prior art documents which were not included in earlier searches, and this is not limited to prior art documents that were published after the earlier search. Furthermore, during litigation concerning the validity / invalidity of a given patent, new prior art information is often found which were missed by the patent office.

[0010] If a patent is a ‘business asset’, which is practically acquired from the government, then it should be a reliable basis for investment and business, almost as reliable as a government bond. Allowing relevant prior art information to escape the process before grant, then be captured after a patent has been granted, and during legal challenges and litigation that can possibly result in an invalidation of the patent is a situation that should never be allowed. This is analogous to investors purchasing land from the government, building a factory on it, raising further funding to set up production lines and a business etc. only to be told that the land purchasing contract is invalid, and that the land is now free for all to use. Prior art searches should be done, exhaustively and on a once-and-for-all basis from the outset, with the possibility of finding previously un-captured relevant information being nothing other than a rare force majeure. It should be noted that prior art information ultimately includes patents, patent applications, utility models, scientific literature and in some cases oral recordings. As such, merely relying on patent documents would inherently leave out potentially crucially important relevant information.

[0011] The other side of the search coin so to speak, is the problem of cluttering irrelevant information. Here the searches may capture prior art information, which whilst may be relevant as a technological background, it is actually irrelevant to the novelty and inventive step of the given invention, which can lead to an unwarranted surge in prosecution efforts and in disputes.

[0012] The present disclosure utilizes the theoretical-practical tools of ‘HalfCausation’, to present a clear-cut logical classification framework, before embedding it in a highly efficient computer-implemented method for searching prior art information sources, and reconstituted patent databases.

[0013] Half-Causation was originally developed as a ‘logical’ model for the dynamic systemization of the process of inventing. It consists of five phase of reasoning, through which technological inventions can be developed. Each phase of reasoning terminates with taking a logical branch. As such, Half-Causation debunks the myth that inventing is mysterious and cannot be systemized.

[0014] SUMMARY OF THE INVENTION

[0015] However, it was only very recently that novel developments regarding the use of Half-Causation for the purpose of efficient classification of prior art information sources, which gave rise to the present disclosure. First, here is an introduction to the specific aspects of Half-Causation, which are relevant to the present patent application. Whilst HalfCausation as an empirical model is based on current patent laws and practices, its theoretical toolbox sometimes introduces theoretically important and practically useful refinements, which might (or might not) be adopted in future legislations and executive procedures.

[0016] Certain aspects or embodiments may reduce or obviate certain problems associated with classifying information, for example, patent documents.

[0017] In a first aspect, a computer-implemented method of classifying invention- related information in relation to one or more sources is provided.

[0018] The method comprises: Identifying an indication of a Distinctive Causal Output, DCO, and one or more Essential Causal Features, ECFs, in invention-related information, wherein the DCO and one or more ECFs are identified using one or more machine learning models trained to identify one or more linguistic key terms in the invention- related information indicative of the DCO and the one or more ECFs.

[0019] The method further comprises: Classifying the invention-related information according to whether or not the indication of the DCO and one or more ECFs indicates that the invention-related information belongs to either: a first class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are different to one or more ECFs of the one or more sources; a second class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are the same as one or more ECFs of the one or more sources. The method further comprises: Providing an indication of whether the invention-related information has been classified as belonging to the first class, the second class or does not belong to either class.

[0020] In a second aspect, a computer-implemented method of modifying data in a database is provided.

[0021] The method comprises: Receiving one or more sources.

[0022] The method further comprises: Classifying the one or more sources according to the computer- implemented method of the first aspect or any related embodiment.

[0023] The method further comprises: Generating an indication of the DCO and the one or more ECFs for the source, and optionally, the DCI and / or one or more EPCs for the source, wherein the indication represents data to be added to the database and / or data to be used to modify existing data in the database.

[0024] In a third aspect, a computer- implemented method of training a machinelearning model for implementing the computer-implemented method of any of the first aspect, second aspect or any related embodiment is provided.

[0025] The method comprises: Receiving one or more sources of a first dataset.

[0026] The method further comprises: Using a first machine learning technique to annotate the one or more sources of the first dataset to indicate the DCO and the one or more ECFs for the source, and optionally, the DCI and / or one or more EPCs for the source.

[0027] In a fourth aspect, an apparatus comprising a processor configured to perform the method of any of the first aspect, second aspect, third aspect or any related embodiment is provided.

[0028] In a fifth aspect, a computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any of the first aspect, second aspect, third aspect or any related embodiment is provided.

[0029] In a sixth aspect, a machine-readable medium comprising instructions which, when executed by a processor, cause the processor to carry out the method of any of the first aspect, second aspect, third aspect or any related embodiment is provided.

[0030] Certain aspects or embodiments described herein may provide various technical benefits such as described herein.

[0031] The present disclosure may provide improved classification and searching of sources such as patent prior art documents, included in both historical databases and ongoing new patent applications which automatically become prior art documents for subsequent patent applications. This is achieved using an electronically -implemented causal language model that is ultimately based on the Half-Causation model of the invention process. The classification and subsequent searches can be made at different levels of causal accuracy and semantic fields so as to suit legislation and executive procedures as they may evolve over the next two decades or so. The invention can provide exhaustive search results, which can furthermore be utilized in determining the existence of an Inventive Step for a given invention, and also the Freedom to Operate. The exhaustiveness of the search results / classification has economic and environmental benefits. Economically, using such a classification approach may reduce, minimize and / or eliminate the risk of the revocation / invalidation of granted patents due the subsequent resurfacing of relevant prior art that was missed before grant, thus reinforcing patents as reliable business assets analogous to government bonds. Environmentally, eliminating the need for repeated searches for the same patent application over and over again especially internationally, can save circa 50 million kWh over 20 years. A further classification step may be performed in respect of sources published after 18 months (the typical confidentiality period for patent applications). However, the range of time over which the classification is to be performed is much less than the complete bulk of patent applications that are electronically available on the internet.

[0032] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiment(s) described hereinafter.

[0033] BRIEF DESCRIPTION OF DRAWINGS

[0034] Exemplary embodiments of the invention will now be described, by way of example only, with reference to the following drawings, in which:

[0035] FIG. 1 refers to a method of classifying information according to an embodiment;

[0036] FIG. 2 refers to a method of classifying information according to an embodiment;

[0037] FIG. 3 refers to a method of modifying data according to an embodiment;

[0038] FIG. 4 refers to a method of training a machine learning model according to an embodiment;

[0039] FIG. 5 refers to a method of training a machine learning model according to an embodiment;

[0040] FIG. 6 is a schematic drawing depicting an approach for classifying information according to an embodiment; FIG. 7 is a schematic drawing depicting an approach for classifying information according to an embodiment;

[0041] FIG. 8 refers to a method of classifying information according to an embodiment;

[0042] FIG. 9 is a schematic drawing of a machine-readable medium for implementing various embodiments; and

[0043] FIG. 10 is a schematic drawing of apparatus for implementing various embodiments.

[0044] DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] A ‘technological invention’ is defined according to the Half-Causation principle [quoted from both Abolkheir 2019 & Abolkheir 2021]:

[0046] Abolkheir, M. (2019) If You Wish to Invent Then Follow the Half-Causation Method. Techne: Research in Philosophy and Technology 23(1): 26-50.

[0047] Abolkheir, M. (2021): How Half-Causation can enlighten the drafting of patent claims. Queen Mary Journal of Intellectual Property. Vol. 11 No. 4, pp. 469-483.

[0048] Thus, a technological invention is a human-made match between two halves of a causal relation, such that:

[0049] One half is a technologically available Distinctive Causal Input [DC Input (DCI)];

[0050] The other half is a technologically desired Distinctive Causal Output [DC Output (DCO)]

[0051] And the match is essentialized by an arrangement of Essential Physical Conditions, which confirms that the match actually works.

[0052] Official guidelines produced by patent offices usually highlight the difference between ‘essential features’ and ‘non-essential’ features. Half-Causation follows that, but then proposes an important split within essential features into: Distinctive Causal Input (DCI), and Essential Physical Conditions (EPCs). The following quote from Abolkheir 2021 illustrates this well: here are the four questions which patent drafters should answer before drafting patent claims:

[0053] 1. “What is the desired Distinctive Causal Output [DC Output] which the invention achieves? I.e. what exactly does the invention purport to achieve?”

[0054] 2. “What is the invention’s Distinctive Causal Input [DC Input]? I.e. what exactly is unique about it?” 3. “What are the invention’s Essential Physical Conditions (EPCs)? I.e. what are the absolute minimum features without which the invention can NOT work? These are to be included in Claim 1 (and any other independent claims e.g. if Claim 1 protects the invention as a device, another independent claim may protect it as a method).”

[0055] 4. “What are the invention’s Accessory Physical Conditions (APCs)? I.e. what are the different variations and alternative features, which should NOT be included in Claim 1 (and any other independent claims), but should be introduced gradually, individually and in combinations, in dependent claims?”

[0056] Throughout this disclosure, reference is made to ‘claims’ (e.g., the reference to ‘claim 1’ above). Depending on the context, such references should not be interpreted as referring directly to the claims in the set of claims appending to this description. Instead, such references are intended to explain the how the principles described herein may be used to analyze patent documents that contain claims.

[0057] The forgoing introductory paragraphs presented the technological problem of prior art searching, and provided a brief introduction to the theoretical-practical tools of HalfCausation. Specifically, it highlighted the three definitions which are utilized in the present disclosure, namely:

[0058] Distinctive Causal Input [DC Input].

[0059] Distinctive Causal Output [DC Output].

[0060] Essential Physical Conditions [EP Conditions].

[0061] An approach of the present disclosure is that the relevance of a prior art information source (an example of invention-related information) is determined on the basis of whether it contains references to a DC Input, a DC output, and in some cases also EP Conditions. Any component or device can be the DC Input in some invention or another. Similarly, there are practically infinite references to desired effects, any of them can be the DC Output in some invention or another. For example, a magnetron is a device that would be routinely mentioned in engineering textbooks and dictionaries. Similarly, references to cooking would be routinely made in culinary cookbooks and magazines. However, it is only if the same information source contains a reference to a magnetron which would be understood as a ‘DC Input’, in relation to what would be understood as the ‘DC Output’ of cooked food that the information source becomes relevant to a prior art searches that relate to cooking methods and devices. As such, it would be practically useless to pay attention to an information source which only contains a reference to some available DC Input, or some desired DC Output. It is only if the same information source contains references to both a DC Input and a related DC Output that it becomes relevant to prior art searches. On that basis, colossal information sources including patent databases and scientific literature etc. can now have a ‘logical structure’, which facilitates them being searched for invention-related prior art searches, potentially with unprecedented efficiency and speed.

[0062] It is to be noted that according to Half-Causation, merely matching a DC Input with a DC Output does not constitute a ‘technological invention’, as the match would only constitute the inventive concept: the idea of the invention, so to speak. For there to be a technological invention, there is also a reference to the EP Conditions. The need for there being EP Conditions [EPCs] is consistent with current practices in terms of demanding there being sufficiency / enablement and, in the USA, reduction to practice (actual or constructive).

[0063] It could also be added that the Half-Causation approach has strong parallels to the EPO’s problem-solution approach, such that the methodological reasoning could start from the problem, i.e., a desired DCO, and then reason towards a matching DCI, or vice versa, start from a solution, i.e., an available DCI, and then reason towards to a matching DCO, following which the EPCs are stipulated to confirm how the invention works.

[0064] FIG. 1 refers to a computer-implemented method 100 of classifying information according to an embodiment. The method 100 may be implemented by a processor such as described in more detail herein.

[0065] The method 100 classifies invention-related information in relation to one or more sources.

[0066] The method 100 comprises, at block 102, identifying an indication of a Distinctive Causal Output, DCO, and one or more Essential Causal Features, ECFs, in invention-related information, wherein the DCO and one or more ECFs are identified using one or more machine learning models trained to identify one or more linguistic key terms in the invention-related information indicative of the DCO and the one or more ECFs;

[0067] The method 100 further comprises, at block 104, classifying the invention- related information according to whether or not the indication of the DCO and one or more ECFs indicates that the invention-related information belongs to either: a first class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are different to one or more ECFs of the one or more sources; or a second class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are the same as one or more ECFs of the one or more sources.

[0068] The method 100 further comprises, at block 106, providing an indication of whether the invention-related information has been classified as belonging to the first class, the second class or does not belong to either class. Such an indication may be textual or some other way of identifying the classification such as a reference to whether the invention is novel and / or inventive, and / or any relevant sources such as an identity of any prior art documents (e.g., patent numbers), as described below.

[0069] Searching for prior art during the patent process is a time-consuming process that consumes energy by submitting queries to structured databases. These queries occur frequently throughout the patent application process by multiple patent offices. This is inefficient and causes significant energy to be used in the process. The wasted energy is significant.

[0070] A Freedom of Information Request submitted to the United Kingdom Intellectual Property Office (UKIPO) yielded for the following information regarding how long searches take. For instances where a time for searching was recorded, the following times relating specifically to searching:

[0071] Average time searching a first search action: 8 hours 15 minutes (1590 first search actions).

[0072] Average time searching a subsequent action: 3 hours 7 minutes. (775 subsequent actions).

[0073] Total time searching 2365 actions recorded in the study: 15549 hours 35 minutes.

[0074] It is thus clear that a significant amount of time and energy is spent performing searching, especially when considering the number of patent applications filed each year, which is of the order of millions.

[0075] Embodiments described herein may reduce the amount of time and energy spent performing searching sources. This extends to other scenarios and is not restricted to searching patent documents. However, for the purposes described herein, we will refer to patent documents.

[0076] Embodiments described herein (including method 100 and related embodiments) may allow for rapid classification of invention-related information (e.g., a patent publication, an invention disclosure form, a draft patent application, a technical note, etc.) with respect to one or more sources (e.g., documents which may have already been classified according to the principles described herein). This point is supported by the calculations provided below:

[0077] Table 1: Calculations of estimated energy costs used when performing prior art searches using traditional methods (e.g., querying patent databases).

[0078] Table 2: Calculations of estimated energy costs used when performing prior art searches using the principles described in the present disclosure.

[0079] Thus, there is a clear reduction in the amount of energy used when using a trained model (as trained according to the principles described herein).

[0080] Of course, there is an energy cost to training. Some estimates are given below in Table 3.

[0081] Table 3: Estimated energy cost for training a model according to the principles described herein. As per the final calculations, models trained in accordance with the present disclosure (i.e., the ‘causal model’) use less energy (e.g., in terms of kWh) over one year than the ‘traditional model.’

[0082] It is therefore apparent that certain embodiments described herein may save significant amounts of energy when performing document searching and classification, as compared to present approaches. This may make it easier for humans to spend more time analyzing the results, which may be more accurate and comprehensive than existing results. Once a set of sources has been classified in accordance with principles described herein, there is no need to perform additional searching with respect to available sources, except after the 18 month confidential period for patent applications.

[0083] Some further embodiments relating to the method 100 are now described. Any one embodiment may be combined with any other embodiment, as appropriate.

[0084] In some embodiments, the method 100 comprises identifying a Distinctive Causal Input, DCI, in the information, based on a second set of one or more linguistic key terms that describe the DCI.

[0085] In some embodiments, the one or more linguistic key terms includes any possible synonyms for the one or more linguistic key terms.

[0086] In some embodiments, classifying the invention-related information comprises classifying the invention-related information into one of a plurality of classes, wherein the plurality of classes comprises the first class, the second class, a third class and a fourth class, wherein the first class and the second class contain one or more of sources that mention one or more of the key terms that relate to the DCO, wherein the third class contains one or more sources that mention one or more key terms that relate to Distinctive Causal Input, DCI, and wherein the fourth class comprises an intersection of the first and second classes with the third class.

[0087] In some embodiments: the DCO is indicative of what the invention purports to achieve; and the one or more ECFs represent one or more features and / or functionality that, when combined, provide the DCO.

[0088] In some embodiments, the invention-related information comprises data representative of a description of an invention.

[0089] FIG. 2 refers to a computer-implemented method 200 of classifying information according to an embodiment. The method 200 may be implemented by a processor or other computing apparatus such as described in more detail herein. The method 200 is dependent on and includes the method 100.

[0090] The method 200 comprises, at block 202, identifying an indication of a Distinctive Causal Input, DCI, and / or one or more Essential Physical Conditions, EPCs, wherein the DCI and one or more EPCs are identified using one or more machine learning models trained to identify one or more linguistic key terms in the invention-related information indicative of the DCI and the one or more EPCs. In this context, an indication may be a pointer (e.g., location in the document) to a portion of the text that comprises the linguistic key term, or the text itself.

[0091] The method 200 further comprises, at block 204, classifying the invention- related information according to whether the indication of the DCI and / or one or more EPCs indicates that the invention-related information belongs to either:

[0092] A first sub-class in which the invention-related information has the same DCI as one or more of the sources; or

[0093] A second sub-class in which the invention-related information has the same DCI as one or more of the sources and, additionally, has the same one or more EPCs as the one or more sources.

[0094] The method 200 further comprises, at block 206, providing an indication of whether the invention-related information has been classified as belonging to the first subclass, the second sub-class or does not belong to either sub-class. Such an indication may be textual or some other way of identifying the classification such as a reference to whether the invention is novel and / or inventive, and / or the identity of any relevant sources such as prior art documents.

[0095] Some embodiments relating to the method 100 and / or method 200 are described below.

[0096] In some embodiments, the DCI is indicative of features and / or functionality that are unique for the invention; and the one or more EPCs are indicative of features and / or functionality that are not unique but without which the invention will not work.

[0097] In some embodiments, the DCI and one or more EPCs represent a split of the one or more ECFs.

[0098] In some embodiments: the DCI is isolated from within the one or more ECFs; or the DCI is identified by comparison to other sources. In some embodiments, if the invention does not belong to a class or sub-class, the provided indication indicates that the invention -related information comprises novel and / or inventive subject-matter.

[0099] In some embodiments, classification of the invention -related information as belonging to one or more classes or sub-classes is indicative of a level of novelty and / or inventiveness of the subject-matter, wherein classification of the invention-related information as not belonging to fewer classes or sub-classes is indicative of a higher level of novelty and / or inventiveness of the subject-matter compared with invention-related information that belongs to relatively more classes or sub-classes.

[0100] In some embodiments, at least one set of linguistic key terms can also include any possible hyponyms, and / or coordinates, and / or hypernyms.

[0101] In some embodiments, the invention-related information comprises a patent document, and wherein classification of the patent document indicates the DCO and the one or more ECFs for the patent document, and optionally, the DCI and / or one or more EPCs for the patent document.

[0102] In some embodiments, the classification indicates the DCO and the one or more ECFs for the patent document, and optionally the DCI and / or one or more EPCs, for the patent document via one or more normalized indicators representative of language data for the DCO, the one or more ECFs, the DCI and / or the one or more EPCs.

[0103] In some embodiments, the one or more linguistic key terms are extracted from one or more independent claims of the patent document, and optionally extracted from one or more dependent claims of the patent document.

[0104] In some embodiments, the invention-related information is received as part of a query to determine whether or not subject-matter in the invention-related information is: novel; inventive; and / or enjoys freedom to operate with respect to one or more sources that have been classified by the one or more machine learning models, and wherein the provided indication indicates whether or not specified subject-matter in the invention-related information is: novel; inventive; and / or enjoys freedom to operate.

[0105] In some embodiments, the invention-related information is received as part of a training dataset for training one or more of the machine learning models, and wherein the invention-related information is used to train the one or more machine learning models.

[0106] In some embodiments, the provided indication indicates an identity of the one or more sources that have the same DCO and one or more ECFs as the invention-related information, and optionally, the provided indication indicates an identity of the one or more sources that have the same DCI and / or one or more EPCs as the invention-related information.

[0107] FIG. 3 refers to a computer-implemented method of modifying data according to an embodiment. The method 300 may be implemented by a processor or other computing apparatus such as described in more detail herein.

[0108] The method 300 refers to modifying data in a database such as an electronic database storing information for classifying a set of sources in accordance with principles described herein.

[0109] The method 300 comprises, at block 302, receiving one or more sources.

[0110] The method 300 further comprises, at block 304, classifying the one or more sources according to the computer-implemented method 100, 200 or any related embodiments;

[0111] The method 300 further comprises, at block 306, generating an indication of the DCO and the one or more ECFs for the source, and optionally, the DCI and / or one or more EPCs for the source, wherein the indication represents data to be added to the database and / or data to be used to modify existing data in the database.

[0112] FIG. 4 refers to a computer-implemented method 400 of training a machine learning model according to an embodiment. The method 400 may be implemented by a processor or other computing apparatus such as described in more detail herein.

[0113] The method 400 trains a machine-learning model for implementing the computer- implemented method 100, 200 or any related embodiments.

[0114] The method 400 comprises, at block 402, receiving one or more sources of a first dataset; and

[0115] The method 400 comprises, at block 404, using a first machine learning technique to annotate the one or more sources of the first dataset to indicate the DCO and the one or more ECFs for the source, and optionally, the DCI and / or one or more EPCs for the source.

[0116] In some embodiments relating to the method 400, an unsupervised learning technique is used to extract one or more linguistic key terms for the DCO and the one or more ECFs for the source, and optionally, the DCI and / or one or more EPCs for the source.

[0117] FIG. 5 refers to a computer-implemented method 500 of training a machine learning model according to an embodiment. The method 500 may be implemented by a processor or other computing apparatus such as described in more detail herein. The method 500 depends on and includes method 400. The method 500 comprises, at block 502, receiving one or more sources of a second dataset that is larger than the first dataset; and

[0118] The method 500 further comprises, at block 504, using a second machine learning technique to train the one or more machine learning models from the one or more sources of the second dataset, wherein the annotated one or more sources of the first dataset are used to supervise the second machine learning technique.

[0119] Some further embodiments that relate to any of the methods 100, 200, 300, 400 and 500 are now provided.

[0120] In some embodiments, the source comprises one or more claims of a patent document.

[0121] In some embodiments, selection of one or more sources is done at random.

[0122] In some embodiments, the selection of the starting sources is determined by the user. For example, the user performing a query via an interface to a system for implementing the embodiments described herein.

[0123] In some embodiments, there is a third set of linguistic key terms for Essential Physical Conditions, which can either be entered by the user or obtained from the source, which can include any possible synonyms, hyponyms, coordinates, and hypernyms.

[0124] In some embodiments, the linguistic key terms that are isolated from the sources are obtained from specifications of patents and patent claims, optionally the independent claims in each patent or patent application, and further optionally any examination report that is available for each patent or patent application.

[0125] In some embodiments, the sources are obtained from patents and patent applications, which have been classified according to an official classification.

[0126] In some embodiments, the sources include audio sources which have been converted into text.

[0127] FIG. 6 is a schematic drawing depicting an approach for classifying information according to an embodiment.

[0128] FIG. 6 is a Venn Diagrams, which shall be referred to as an Abolkheir Standard Inventiveness Diagram [ASID]. ASIDs depict an approach for classifying information according to an embodiment. ASIDs are intended to provide the logical and methodological basis of the classification of prior art documents, and subsequent prior art searches. Furthermore, ASIDs can assist with judging the existence of an Inventive Step for a given invention, and / or the existence of Freedom to Operate. The population of ASIDs are ‘referenced causations’ i.e. it is not the prior art documents themselves (the information sources) that populate the diagrams, but the causal processes that they refer to. Throughout the next few paragraphs, the invention of the microwave oven (US 2,495,429) will be used as a (simplified) example of how ASIDs work. To avoid repetition, all new terms will be introduced in the text for the following figures, and thereafter only new terms will be introduced as required.

[0129] FIG. 6 represents a first ASID, ASID-1, which aims to provide a macro picture of different aspects of causal processes, before gradually zooming in on the more essential parts of relevant causal processes. It is helpful to note that it is the DCO that sets the relevance for the input side of any causal process. ASID-1 includes the following terms and abbreviations:

[0130] DCO: Distinctive Causal Output, which is the desired output of the causal process i.e. what the invention purports to achieve. In the microwave oven example, that would be ‘cooked food’.

[0131] Aspirational: DCO as referenced in non-technical sources e.g. newspapers, magazines and social media. The references here would be mostly non-technical, indicating for example how desirable it would be to have a speedy cooking method that suits modem / contemporary busy and rapid lifestyles.

[0132] APCI: All Possible Causal Input, as referenced in engineering and technical sources, and also includes patent documents. Basically, this encompasses references to vast causal inputs such as components e.g. magnetrons, tubes etc. some of which may be matched with one DCO or another, others may be referenced without being matched with a DCO.

[0133] In Textbooks: Only referenced in textbooks and dictionaries etc. but without being matched with a DCO.

[0134] Given DCO-ECFs Match: DCO matched with one set of Essential Causal Features (all causal features in independent claims in a given patent document). In the microwave oven example, Essential Causal Features in Claim 1 include (using simplified terminology): microwave energy, restricted region of space [enclosure], and concentrating and guiding means.

[0135] Other DCO-ECFs Matches: The same DCO but matched with other and different sets of Essential Causal Features (each consists of all causal features in independent claims in respective patent documents). These would be numerous other cooking methods and devices. FIG. 7 is a schematic drawing depicting an approach for classifying information according to an embodiment. FIG. 7 is another Venn diagram in the form of an ASID.

[0136] FIG. 7 represents a second ASID, ASID-2, which is an advanced version of ASID-1. The difference is that ECFs (Essential Causal Features: all causal features in independent claims in a given patent document) are split into DCI: Distinctive Causal Input, which is what is unique about the invention, and EPCs: Essential Physical Conditions - which are not unique but without them the invention will not work, and as such are also mentioned in the independent claims. [Note: Splitting ECFs into DCI and EPCs is a logical and methodological development introduced by the Half-Causation model of the inventive process] This split is illustrated by introducing a further subset (circle), which is marked EPCs. [Note: knowledge of what the DCI is, is ‘epistemically prior’ to knowledge of what EPCs are relevant to it i.e. the DCI must be know first, then it would be possible to know what EPCs would work with it. That is why EPCs are illustrated as a subset of DCI] So, in the microwave oven example, what is unique about the invention is precisely the use of microwave energy as the DCI, which is matched with the DCO of cooked food. The enclosure and the concentrating and guiding means etc. apply to the causal process and that is why they are stated in Claim 1, but they are not on par with the DCI, which plays a unique role in the causal process. The splitting of the ECFs into DCI and EPCs may have significant advantages when undertaking: classification and searching of prior art documents, and furthermore when judging the existence of an Inventive Step and Freedom to Operate.

[0137] The ASIDs, and corresponding classifications and searches can be constructed at different semantic fields. In simple terms, the filed could be set at: Same / Synonyms / Hyponyms [SSH], or it could be set up at: Hypernym [Hr]. So, continuing with the microwave oven example, if the search is conducted at SSH, then the term ‘microwave energy’, which refers to the DCI, would be normalized to include, for example, ‘electromagnetic wave energy of a wavelength falling in the microwave region of the electromagnetic spectrum’ (synonym), and ‘electromagnetic waves with a frequency of 2450 MHz’ (hyponymy). As such, a search at the SSH level / field for DCIs which are matched with the DCO, captures all prior art documents in which the DCO is matched with DCIs expressed in any terms within the SSH filed.

[0138] In contrast to current search approaches that use one or a combination of traversal of a classification and keyword searches, the present disclosure uses the duplet of DC Input and DC Output to improve the recall and precision of search. In a preferred embodiment, the present disclosure uses the triplet of DC Input, DC Output, and EP Conditions. By using the triplet the results of the search bound is a much greater space of recall and hence potentially relevant prior art. Similarly, results that lie at the intersection of the three sets of results represents a highly precise and likely highly relevant prior art. As such, the present disclosure is expected to achieve the highest Fl score.

[0139] It is useful to highlight that the present disclosure is not limited to current legislation, case law and executive procedures, any and all of which may evolve over the next years, especially in the light challenges posed by advances in Artificial Intelligence. As such, the user is able to set the criteria in the present disclosure for the classification and / or searches according to the prevailing legislation, case law, and executive procedures at the time.

[0140] Based on the approaches described in the present disclosure, any user anywhere in the world would be able to find some prior art document somewhere which mentions a DCO, find another that mentions causal input that could act as a DCI, and perhaps also causal input that could act as EPCs, and as such raise an objection as a third party arguing that the given invention would have been obvious. It is envisaged that the DCO, DCI and EPCs for a prior art document (as determined in accordance with principles described herein) could be referenced in the same prior art document (information source).

[0141] Some useful contextual information on how the Inventive Step can be defined is now provided.

[0142] The distinction between an invention having or not having an inventive step is based on whether the notional person with their unimaginative powers would have or would have not interpreted and used prior art information to clinch the core inventive concept. The objective of patent law is to examine the invention itself in relation to prior art inventions. This understanding is clear from the fact that prior to the UK 1932 Act the focus of the examination was on novelty i.e. on the outcome of a comparison between invention information and prior art information.

[0143] If a decision is made that an inventive step exists between invention information and prior art information, then the decision implies an assertion that the notional person lacked the required imaginative cognitive capacities to clinch the invention’s inventive concept (‘the given inventive concept’). This assertion implies that there is a threshold of imaginative cognitive capacities, below which a person cannot clinch the given inventive concept. As such, having imaginative cognitive capacities above the threshold is an implied necessary (but insufficient) condition for a person to clinch the given inventive concept. So, asserting that the notional person’s imaginative cognitive capacities were (at least on this occasion) below the threshold is logically equivalent to asserting that the actual inventor’s imaginative capacities were (at least on this occasion) above the threshold. In effect, patent granting authorities have been unwittingly examining the cognitive capacities of actual inventors and asserting that they have imaginative capacities.

[0144] Deploying and using the principles described in the present disclosure provides an objective approach to determining whether or not invention -related information is novel and / or inventive, and / or whether there is freedom to operate in respect. This objective approach may be considered to be consistent with or at least complement the current approach of patent offices and the courts when assessing novelty, inventive step and / or freedom to operate. In other similar words, the present disclosure provides technological tool to assist with determining the ‘distance’ between a claimed invention and the relevant prior art.

[0145] The present disclosure uses the following terms: Distinctive Causal Input (DCI), Distinctive Causal Output (DCO), and Essential Physical Conditions (EPCs). These terms ultimately refer to concepts, as follows. As mentioned above, patent practices usually highlight the ‘essential features’ of a given invention. However, Half-Causation introduces a split within these features into: Distinctive Causal Input, and Essential Physical Conditions. This is a conceptual split, and not merely linguistic. Similarly, Distinctive Causal Output refers to a special part of the output of the causal process. For example, when a raw chicken is cooked in a microwave oven, we could refer to the cooked chicken as the desired Distinctive Causal Output. But it is not the only the output. Other outputs include, for example, the scattered microwave emissions which were not absorbed by the raw chicken, increased heat inside the cavity (enclosure) within which the cooking takes place, and the consumed electricity etc. Once these concepts are understood, then numerous different linguistic terms can be used to refer to them. Linguistic terms (which may also be referred to as linguistic key terms) refer to features and / or functionality that define an invention and how it works. For example, the term ‘Distinctive’ can be replaced by: unique, distinguished, selected, highlighted, singular, individual, characteristic etc. Similarly, the term ‘Causal’ can be replaced by: causational, functional, operational etc. or it may not be used at all e.g. instead of Distinctive Causal Input we could say: unique causer, or distinguished inducer etc.; and instead of Distinctive Causal Output we could say: unique outcome, or teleological end etc. The term Output can be replaced by: effect, outcome, result, objective, purpose, aim, goal state, end state, intended end, final cause, telos etc. The term Essential can be replaced by: fundamental, indispensable, irreplaceable etc. The term Physical can be replaced by: non-mental, material etc., and furthermore it can link back to: causal, functional, operational etc. The term Conditions can be replaced by: features, stipulations, provisions, circumstances etc. So, the terms Distinctive Causal Input, Distinctive Causal Output, and Essential Physical Conditions, as used in the present disclosure should be understood as the highest-order linguistic terms, each referring to a concept which may be referred to by any number of alternative linguistic terms, be they synonyms or hyponyms.

[0146] One of the features of the present disclosure (or to use Half-Causation’s own terminology: one of its Essential Physical Conditions) is the use of lexical semantics to capture prior art information sources. Natural Language Processing [NLP] is used in machine learning techniques described herein. When applying lexical semantics, attention needs to paid to synonyms, and preferably and as appropriate to hyponyms, coordinates and hypernyms.

[0147] Here is an example of a semantic tree or ramified hierarchy from mechanical engineering. The term ‘mechanical fixing’ is synonymous to e.g. ‘fixing & fasteners’. Within it we could have ‘permanent mechanical fixing’ which is synonymous with ‘irreversible mechanical fixings’, both of which would be coordinates of ‘non -permanent mechanical fixing’ which is synonymous with ‘reversible (or adjustable) mechanical fixing’. Within ‘permanent mechanical fixing’ we could have ‘rivet’ which is synonymous with ‘stud’, both of which would be coordinates of ‘nail’ which is synonymous with ‘tack’. Within ‘nonpermanent mechanical fixing’ we could have ‘bolt’ which is synonymous with ‘screw’, both of which would be coordinates of ‘clamp’ which is synonymous with ‘press’. And so on.

[0148] The following is an embodiment of the present disclosure, which is provided as an example: a computer-implemented method for searching prior art information sources that relate to a given invention, comprising:

[0149] Instructions to prompt the user to enter the given invention’s Distinctive Causal Input, and its Distinctive Causal Output, using linguistic key terms;

[0150] Instructions to form a set of linguistic key terms, one for the the Distinctive Causal Input, and the other for the Distinctive Causal Output, with each set, respectively, including all the linguistic key terms that describe the Distinctive Causal Input and the Distinctive Causal Output, which besides the same key terms used by the user also includes any possible synonyms.

[0151] Instructions to use the two sets of linguistic terms to conduct a search of patent data bases and any other relevant sources of prior art information, to identify the sources which include references to both the Distinctive Causal Input and the Distinctive Causal Output, in which any of the linguistic terms from the two sets were used;

[0152] Instructions to present the findings. Optional features include at least one set of linguistic key terms also including any possible hyponyms, and / or coordinates, and / or hypernyms. The method can include instructions to prompt the user to approve the two sets of linguistic terms before conducting the search. Another embodiment is a data processing apparatus, comprising means for carrying out the above-mentioned instructions of the method. And another embodiment is a computer program [product], comprising instructions which, when the program is executed by the computer, cause the computer to carry out the above-mentioned method. Yet another embodiment is a computer readable medium having stored there on the computer the above- mentioned program [product] .

[0153] Another embodiment is a computer-implemented method for the normalization of text in invention-related information sources, which can subsequently be used to identify prior art information sources that relate to a given invention, comprising: Receiving invention- related information; Normalizing the information according to the following steps:

[0154] Instructions to identify the Distinctive Causal Input and the Distinctive Causal Output in each invention-related information source;

[0155] Instructions to isolate the linguistic key terms that describe both the Distinctive Causal Input and Distinctive Causal Output in each source;

[0156] Instructions to form a set of linguistic key terms for the Distinctive Causal Input, and another set for the Distinctive Causal Output, each set containing besides the same key terms used in the source also includes any possible synonyms.

[0157] Instructions to classify the sources into three classes: one containing all sources that mention any of the key terms in the set that relates to the Distinctive Causal Input, another containing all sources that mention any of the key terms in the set that relates to the Distinctive Causal Output, and a third which is the intersection of the two.

[0158] Then Presenting the result based on the said normalized information.

[0159] Optional features include at least one set of linguistic key terms also including any possible hyponyms, and / or coordinates, and / or hypernyms.

[0160] Other embodiments are: data processing apparatus comprising means for carrying out the method mentioned in the previous paragraph; a computer program [product] comprising instructions which when the program is executed by the computer, cause the computer to carry out the said method; and a computer readable medium having stored there on the said computer program [product] . Optional features include the selection of the starting sources being done at random. Alternatively, the selection of the starting sources is determined by the user. Another embodiment is a digital database containing the result of the method presented in this paragraph.

[0161] Further optional features include there being a third set of linguistic key terms for Essential Physical Conditions, which can either be entered by the user or obtained from the source, which can include any possible synonyms, hyponyms, coordinates, and hypernyms.

[0162] The linguistic key terms that are isolated from the sources can be obtained from specifications of patents and patent claims, specifically the independent claims in each patent or patent application plus any examination report that is available for each patent or patent application. These sources could have already been classified according to an official classification e.g. the International Patent Classification [IPC], or the Cooperation Patent Classification [CPC]. In this case, the user can use to the official classification to determine the class or classes and sub-classes in which in their opinion the search should be conducted, then let the present disclosure do its work within the determined research area. However, this approach might unwittingly allow escapes if the original classification of (some) patents and / or patent applications was not accurate.

[0163] The information sources can include audio sources which have been converted into text. The method can include means for translating the language of the original source into at least one other language. The computer readable medium can be downloadable software, or online computing including cloud computing.

[0164] An advanced version of the computer-implemented method is to include instructions to present a judgement on the novelty of the given invention, and possibly also a judgement on the inventive step of the given invention. These judgements can be based on criteria obtained from current legislative and executive practices including case law, which may change from time to time, and which can be entered by the user. The method can include instructions to prompt the user to approve the judgement.

[0165] Any of the above variations can also include instructions to highlight the most relevant part of the text in each information source.

[0166] FIG. 8 refers to a computer- implemented method 800 of classifying information according to an embodiment. The method 800 may be implemented by a processor or other computing apparatus such as described in more detail herein. One or more blocks may be omitted, or performed in a different order to that indicated, when implementing the method 800 in accordance with the various embodiments described herein.

[0167] At block 802 of the method 800, a patent document is obtained. At block 804 of the method 800, the independent claims are extracted from the patent document. In some cases, one or more dependent claims may be extracted as well.

[0168] At block 806 of the method 800, causal keywords in the claim(s) are identified.

[0169] At block 808 of the method 800, DCO and one or more ECFs are identified.

[0170] In some cases, at block 810 of the method 800, the DCI is isolated within the one or more ECFs.

[0171] In some cases, at block 812 of the method 800, the DCI is identified by comparison to other sources (e.g. other patent documents that have been classified).

[0172] Block 810 and block 812 proceed to block 814 of the method 800.

[0173] At block 814, in some embodiments, the one or more ECFs are split into DCI and EPC.

[0174] At block 816 of the method 800, the patent document’s DCO, DCI and one or more EPCs are annotated (or simply, only the DCO and one or more ECFs).

[0175] At block 818 of the method 800, the language of the DCO, DCI and one or more EPCs (or simple, only the DCO and one or more ECFs) is normalized.

[0176] At block 820 of the method 800, the patent document is classified according to the normalized language. The classification may indicate information such as whether or not there are any documents relevant to the novelty, inventiveness, freedom to operate of the idea described by the patent document, and / or the identity of any relevant documents.

[0177] An example of training one or more machine learning models, as per one or more embodiments described herein, is now provided. The training is performed by using patent documents as sources (e.g., published patent applications and / or granted patents, which are referred to simply as ‘patents for ease of reference). This approach is referred to as extracting causal relations from patent documents. The size of the

[0178] The aim is to provide a novel annotation of patent documents for the purposes of relevant document retrieval and a method for implementing this annotation. The novel annotation of patents will be based on identification of three parts:

[0179] Distinctive Causal Output (DCO).

[0180] Distinctive Causal Input (DCI).

[0181] Essential Physical Conditions (EPC).

[0182] It is to be noted that in some embodiments, not all of these parts need to be extracted. For example, in some embodiments, the DCO and one or more Essential Causal Features, ECFs are identified by using one or more machine learning models trained according to similar principles as described below. In some embodiments, the DCI and one or more Essential Physical Conditions (EPCs) represent a split of the one or more ECFs. In such embodiments, the DCI and the one or more EPCs are identified by using one or more machine learning models trained according to similar principles as described below.

[0183] The example below refers to various numbers of documents (of an assumed size) used at different stages of training. It should be understood that these numbers are examples and could be varied as appropriate.

[0184] The scale of the annotation task is in the region of 50m documents, each comprising on average 15 pages, with on average 250 words per page. Minimal labelled data exists for this task. Patents are first annotated with the DCO, DCI, and EPC. After annotation, documents are searchable via keyword retrieval.

[0185] The main annotation task is divided into a pilot annotation task, which provides a set of labelled data, and a full annotation task, which leverages large scale natural language processing systems to annotate the full corpus of documents.

[0186] The pilot task uses unsupervised named entity recognition to annotate approximately 50,000 patent documents. The accuracy of the system is checked by manual inspection of up to 500 documents.

[0187] The full task uses the annotated dataset generated by the pilot task to fine-tune a large-scale language model such as Bidirectional Encoder Representations from Transformers (BERT, Devlin et al., 2019) to annotate the full corpus of around 50m documents.

[0188] Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) (pp. 4171-4186).

[0189] Further details of the Pilot are described below.

[0190] Each patent is annotated with the DCO, DCI, and EPC (noting that in some embodiments not all of these are needed - e.g., only the DCO and one or more ECFs are used in some embodiments). In some implementations, the DCO, DCI, and EPC can be found within the main claims of the patent.

[0191] In a first stage of the pilot, the claims are extracted. The main claims of the patent are extracted from the document by first identifying each claim, and subsequently discarding those that depend on previous claims. To identify each claim, a set of phrases is identified that introduce the list of claims, such as: ‘The claims are as follows:’. Subsequently each paragraph is extracted. If the paragraph makes reference to a previous claim, we discard it. To identify whether a paragraph makes reference to a previous claim, we again identify a list of key words / phrases such as ‘ . . . as in claim . . . ’ .

[0192] In a second stage of the pilot, each patent in the dataset is annotated.

[0193] The identification of the DCO, DCI, and EPC can be viewed as three unsupervised named entity recognition tasks (noting that in some embodiments not all of these are needed - e.g., only the DCO and one or more ECFs are used in some embodiments). Unsupervised named entity recognition is used to extract key words and phrases for each of the three categories. Approaches to identification of the DCI and DCO (or simply, only the DCO and one or more ECFs) include:

[0194] Causal keyword identification: causal keywords such as ‘comprising’, ‘so as to provide’, ‘thus providing’, ‘thus producing’, ‘thus causing’, etc., can be identified in the claims. In the phrase ‘An X comprising Yl, Y2, Y3, such that Z W V’, we extract ‘An X such that Z W V’ as a DCO and Yl, Y2, Y3 as DCI. Causal keywords can be identified as detailed in Yang et al 2022. See Yang, J., Han, S. C., & Poon, J. (2022). A survey on extraction of causal relations from natural language text. Knowledge and Information Systems, 64(5), 1161-1186.

[0195] Causal keyword bootstrapping, as detailed in Etzioni et al., (2005) will be leveraged to identify more causal keywords and additional DCO and DCI. See Etzioni, O., Cafarella, M., Downey, D., Popescu, A. M., Shaked, T., Soderland, S., & Yates, A. (2005). Unsupervised named-entity extraction from the web: An experimental study. Artificial intelligence, 165(1), 91-134.

[0196] A small quantity of expert-labelled documents and causal-keywords labelled documents will be used as input to a supervised learning algorithm for named entity recognition. Supervised learning algorithms to be used include: hidden Markov models, support vector machines, decision trees, or conditional random field networks.

[0197] A corresponding approach may be used for annotating the one or more EPCs

[0198] Following automatic annotation of approximately 50,000 documents, up to 500 are manually checked for quality assurance purposes. The annotation with DCI and DCO (or simply, only the DCO and one or more ECFs) will be compared with the International Patent Office classification system to identify any substantial differences or outliers.

[0199] Details of the main training task is now described.

[0200] In a first stage of the main task, the claims are extracted from patent documents as in the first stage of the pilot study. In a second stage of the main task, the dataset of 50,000 documents annotated in the pilot are used as training data for a larger-scale supervised learning approach to named entity recognition. A range of supervised learning approaches can be used, as explained below.

[0201] The supervised learning approaches used in the pilot are extended to the full set of patent documents to identify DCI, DCO, and EPC (or simply, only the DCO and one or more ECFs).

[0202] Deep neural architectures such as Long Short-Term Memories (LSTMs), BERT, Generative Pre-training Transformer (e.g., GPT-3, GPT-4, etc.) are fine-tuned to identify DCI, DCO, EPC (or simply, only the DCO and one or more ECFs).

[0203] Output quality is checked by comparing the output of different models. Where multiple models agree on DCI, DCO, EPC, the patent is annotated with the relevant keywords / phrases. Comparison with the International Patent Office classification may be made to identify outliers.

[0204] FIG. 9 is a schematic drawing of a machine-readable medium 900 (which may or may not be a non-transitory machine-readable medium) for implementing various embodiments described herein. As used herein, the term “non-transitory” does not encompass transitory propagating signals. The machine-readable medium 900 stores instructions 902 readable and executable by a processor 904 to implement the method of any of the embodiments described herein (e.g., method 100, 200, 300, 400, 500 and / or related embodiments). The machine-readable medium 900 and / or the processor 904 may be implemented by any type of computer or processing apparatus, such as described herein.

[0205] FIG. 10 is a schematic drawing of apparatus 1000 for implementing various embodiments described herein. The apparatus 1000 may be implemented by any type of computer or processing apparatus, such as described herein..

[0206] The apparatus 1000 comprises a processor 1002. The processor 1002 is configured to communicate with an interface 1004. The interface 1004 may be any interface (wireless or wired) implementing a communications protocol to facilitate exchange of data (e.g., invention-related information and any other sources) with other devices such as another part of the ecosystem for implementing embodiments described herein.

[0207] The apparatus 1000 further comprises a memory 1006 (e.g., non-transitory or otherwise) storing instructions 1008 readable and executable by the processor 1002 to implement various embodiments described herein (e.g., methods 100, 200, 300, 400, 500 and / or any of the associated embodiments). 1

[0208] Any of the models described herein may be implemented by the processing circuitry for implementing the methods described herein. Thus, certain blocks of the methods may involve use of such models in order to provide the stated functionality. The models may be (machine learning) ML-based or non-ML-based.

[0209] The disclosure includes the subject-matter set out by the following numbered paragraphs.

[0210] 1. A computer- implemented method for searching prior art information sources that relate to a given invention, comprising:

[0211] Instructions to prompt the user to enter the given invention’ s Distinctive Causal Input, and its Distinctive Causal Output, using linguistic key terms;

[0212] Instructions to form a set of linguistic key terms, one for the Distinctive Causal Input, and the other for the Distinctive Causal Output, with each set, respectively, including all the linguistic key terms that describe the Distinctive Causal Input and the Distinctive Causal Output, which besides the same key terms used by the user also includes any possible synonyms.

[0213] Instructions to use the two sets of linguistic terms to conduct a search of patent data bases and any other relevant sources of prior art information, to identify the sources which include references to both the Distinctive Causal Input and the Distinctive Causal Output, in which any of the linguistic terms from the two sets were used;

[0214] Instructions to present the findings.

[0215] 2. A computer- implemented method of Paragraph 1, in which at least one set of linguistic key terms can also include any possible hyponyms, and / or coordinates, and / or hypernyms.

[0216] 3. A computer- implemented method of any preceding paragraph, which also includes instructions to prompt the user to approve the two sets of linguistic terms before conducting the search.

[0217] 4. Data processing apparatus, comprising means for carrying out the method of Paragraph 1.

[0218] 5. A computer program [product], comprising instructions which, when the program is executed by the computer, cause the computer to carry out the method of Paragraph 1.

[0219] 6. A computer readable medium having stored there on the computer program [product] of Paragraph 5. 7. A computer- implemented method for the normalization of text in invention- related information sources, which can subsequently be used to identify prior art information sources that relate to a given invention, comprising:

[0220] Receiving invention-related information;

[0221] Normalizing the information according to the following steps: o Instructions to identify the Distinctive Causal Input and the Distinctive Causal

[0222] Output in each invention-related information source; o Instructions to isolate the linguistic key terms that describe both the

[0223] Distinctive Causal Input and Distinctive Causal Output in each source; o Instructions to form a set of linguistic key terms for the Distinctive Causal

[0224] Input, and another set for the Distinctive Causal Output, each set containing besides the same key terms used in the source also includes any possible synonyms. o Instructions to classify the sources into three classes: one containing all sources that mention any of the key terms in the set that relates to the Distinctive Causal Input, another containing all sources that mention any of the key terms in the set that relates to the Distinctive Causal Output, and a third which is the intersection of the two;

[0225] Present the result based on the said normalized information.

[0226] 8. A computer- implemented method of Paragraph 7, in which at least one set of linguistic key terms can also include any possible hyponyms, and / or coordinates, and / or hypernyms.

[0227] 9. Data processing apparatus comprising means for carrying out the method of Paragraph 7.

[0228] 10. A computer program [product] comprising instructions which when the program is executed by the computer, cause the computer to carry out the method of Paragraph 7.

[0229] 11. A computer readable medium having stored there on the computer program [product] of Paragraph 10.

[0230] 12. A computer-implemented method of Paragraph 7, in which the selection of the starting sources is done at random.

[0231] 13. A computer- implemented method of Paragraph 7, in which the selection of the starting sources is determined by the user.

[0232] 14. A digital database containing the result of the method of Paragraph 7.

[0233] 15. A computer- implemented method of Paragraph 1 or Paragraph 7, which also includes a third set of linguistic key terms for Essential Physical Conditions, which can either be entered by the user or obtained from the source, which can include any possible synonyms, hyponyms, coordinates, and hypernyms.

[0234] 16. A computer-implemented method of either Paragraph 1 or Paragraph 7, in which the linguistic key terms that are isolated from the sources are obtained from specifications of patents and patent claims, specifically the independent claims in each patent or patent application plus any examination report that is available for each patent or patent application.

[0235] 17. A computer-implemented method of Paragraph 1 or Paragraph 7, in which the sources are obtained from patents and patent applications, which have been classified according to an official classification.

[0236] 18. A computer-implemented method of paragraph 17, in which the official classification is the International Patent Classification [IPC] .

[0237] 19. A computer-implemented method of Paragraph 17, in which the official classification is the Cooperation Patent Classification [CPC].

[0238] 20. A computer method of Paragraph 1 or Paragraph 7, in which the sources include audio sources which have been converted into text.

[0239] 21. A computer method of Paragraph 1 or 7, which also includes means for translating the language of the original source into at least one other language.

[0240] 22. A computer readable medium of Paragraph 6 or Paragraph 11, in which the medium is downloadable software.

[0241] 23. A computer readable medium of Paragraph 6 or 11, in which the medium is online computing.

[0242] 24. A computer readable medium of Paragraph 23, in which the online computing is cloud computing.

[0243] 25. A computer- implemented method of Paragraph 1, which also includes instructions to present a judgement on the novelty of the given invention.

[0244] 26. A computer- implemented method of Paragraph 1, which also includes instructions to present a judgement on the inventive step of the given invention.

[0245] 27. A computer-implemented method of Paragraph 25, or Paragraph 26, which also includes instructions to prompt the user to approve the judgement.

[0246] 28. A computer- implemented method of Paragraph 1, or Paragraph 7, which also includes instructions to highlight the most relevant part of the text in each information source. While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The disclosure is thus not limited to the disclosed embodiments.

[0247] One or more features described in one embodiment may be combined with or replace features described in another embodiment.

[0248] Embodiments in the present disclosure can be provided as methods, systems or as a combination of machine -readable instructions and processing circuitry. Such machine- readable instructions may be included on a non-transitory machine (for example, computer) readable storage medium (including but not limited to disc storage, CD-ROM, optical storage, flash storage, or any other appropriate computer storage technology) having computer readable program codes therein or thereon.

[0249] The present disclosure is described with reference to flow charts and block diagrams of the method, devices, and systems according to embodiments of the present disclosure. Although the flow charts described above show a specific order of execution, the order of execution may differ from that which is depicted. Blocks described in relation to one flow chart may be combined with those of another flow chart. It shall be understood that each block in the flow charts and / or block diagrams, as well as combinations of the blocks in the flow charts and / or block diagrams can be realized by machine-readable instructions.

[0250] The machine-readable instructions may, for example, be executed by a general-purpose computer, a special purpose computer, an embedded processor, one or more virtual machines of a cloud-based computing system (e.g., via virtualization), or processors of other programmable data processing devices to realize the functions described in the description and diagrams. In particular, a processor or processing circuitry, or a module thereof, may execute the machine -readable instructions. Thus, functional modules of apparatus and other devices described herein may be implemented by a processor executing machine-readable instructions stored in a memory, or a processor operating in accordance with instructions embedded in logic circuitry. The term ‘processor’ is to be interpreted broadly to include a central processing unit (CPU), processing unit, application specific integrated circuit (ASIC), logic unit, or programmable gate array (including a field programmable gate array (FPGA)) etc. The methods and functional modules may all be performed by a single processor or divided amongst several processors. Such machine-readable instructions may also be stored in a computer readable storage that can guide the computer or other programmable data processing devices to operate in a specific mode.

[0251] Such machine-readable instructions may also be loaded onto a computer or other programmable data processing devices, so that the computer or other programmable data processing devices perform a series of operations to produce computer-implemented processing, thus the instructions executed on the computer or other programmable devices realize functions specified by block(s) in the flow charts and / or in the block diagrams.

[0252] Further, the teachings herein may be implemented in the form of a computer program product, the computer program product being stored in a storage medium and comprising a plurality of instructions for making a computer device implement the methods recited in the embodiments of the present disclosure.

[0253] One or more blocks, elements or steps described in relation to one embodiment may be combined with or replaced by one or more blocks, elements or steps described in relation to another embodiment. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality. A single processor or other unit (such as a virtualized machine of a cloud computing system) may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. A computer program may be stored or distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope of the disclosure.

Claims

CLAIMS:

1. A computer- implemented method (100) of classifying invention-related information in relation to one or more sources, the method comprising:- Identifying (102) an indication of a Distinctive Causal Output, DCO, and one or more Essential Causal Features, ECFs, in invention-related information, wherein the DCO and one or more ECFs are identified using one or more machine learning models trained to identify one or more linguistic key terms in the invention-related information indicative of the DCO and the one or more ECFs;- Classifying (104) the invention-related information according to whether or not the indication of the DCO and one or more ECFs indicates that the invention-related information belongs to either: a first class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are different to one or more ECFs of the one or more sources; or a second class in which the invention-related information has: the same DCO as one or more of the sources; and one or more ECFs of the invention-related information are the same as one or more ECFs of the one or more sources;- Providing (106) an indication of whether the invention-related information has been classified as belonging to the first class, the second class or does not belong to either class.

2. The computer-implemented method of claim 1, comprising identifying a Distinctive Causal Input, DCI, in the information, based on a second set of one or more linguistic key terms that describe the DCI.

3. The computer-implemented method of any of claims 1 to 2, wherein the one or more linguistic key terms includes any possible synonyms for the one or more linguistic key terms.

4. The computer-implemented method of any of claims 1 to 3, wherein classifying the invention-related information comprises classifying the invention-related information into one of a plurality of classes, wherein the plurality of classes comprises the first class, the second class, a third class and a fourth class, wherein the first class and the second class contain one or more of sources that mention one or more of the key terms that relate to the DCO, wherein the third class contains one or more sources that mention one or more key terms that relate to Distinctive Causal Input, DCI, and wherein the fourth class comprises an intersection of the first and second classes with the third class.

5. The computer-implemented method of any of claims 1 to 4, wherein: the DCO is indicative of what the invention purports to achieve; and the one or more ECFs represent one or more features and / or functionality that, when combined, provide the DCO.

6. The computer-implemented method of any of claims 1 to 5, wherein the invention-related information comprises data representative of a description of an invention.

7. The computer-implemented method (200) of any of claims 1 to 6, comprising: Identifying (202) an indication of a Distinctive Causal Input, DCI, and / or one or more Essential Physical Conditions, EPCs, wherein the DCI and one or more EPCs are identified using one or more machine learning models trained to identify one or more linguistic key terms in the invention-related information indicative of the DCI and the one or more EPCs; andClassifying (204) the invention-related information according to whether the indication of the DCI and / or one or more EPCs indicates that the invention-related information belongs to either:A first sub-class in which the invention-related information has the same DCI as one or more of the sources; orA second sub-class in which the invention-related information has the same DCI as one or more of the sources and, additionally, has the same one or more EPCs as the one or more sources;Providing (206) an indication of whether the invention-related information has been classified as belonging to the first sub-class, the second sub-class or does not belong to either sub-class.

8. The computer-implemented method of claim 7, wherein: the DCI is indicative of features and / or functionality that are unique for the invention; andthe one or more EPCs are indicative of features and / or functionality that are not unique but without which the invention will not work.

9. The computer-implemented method of any of claims 7 to 8, wherein the DCI and one or more EPCs represent a split of the one or more ECFs.

10. The computer-implemented method of any of claims 7 to 9, wherein: the DCI is isolated from within the one or more ECFs; or the DCI is identified by comparison to other sources.

11. The computer-implemented method of any of claims 1 to 10, wherein if the invention does not belong to a class or sub-class, the provided indication indicates that the invention-related information comprises novel and / or inventive subject-matter.

12. The computer-implemented method of claim 11, wherein classification of the invention-related information as belonging to one or more classes or sub-classes is indicative of a level of novelty and / or inventiveness of the subject-matter, wherein classification of the invention-related information as not belonging to fewer classes or sub-classes is indicative of a higher level of novelty and / or inventiveness of the subject-matter compared with invention- related information that belongs to relatively more classes or sub-classes.

13. The computer-implemented method of any of claims 1 to 12, in which at least one set of linguistic key terms can also include any possible hyponyms, and / or coordinates, and / or hypernyms.

14. The computer-implemented method of any of claims 1 to 13, wherein the invention-related information comprises a patent document, and wherein classification of the patent document indicates the DCO and the one or more ECFs for the patent document, and optionally when dependent on claim 7, the DCI and / or one or more EPCs for the patent document.

15. The computer-implemented method of claim 14, wherein the classification indicates the DCO and the one or more ECFs for the patent document, and optionally the DCI and / or one or more EPCs, for the patent document via one or more normalized indicators representative of language data for the DCO, the one or more ECFs, the DCI and / or the one or more EPCs.

16. The computer-implemented method of any of claims 14 to 15, wherein the one or more linguistic key terms are extracted from one or more independent claims of the patent document, and optionally extracted from one or more dependent claims of the patent document.

17. The computer-implemented method of any of claims 1 to 16, wherein the invention-related information is received as part of a query to determine whether or not subject-matter in the invention-related information is: novel; inventive; and / or enjoys freedom to operate with respect to one or more sources that have been classified by the one or more machine learning models, and wherein the provided indication indicates whether or not specified subject-matter in the invention-related information is: novel; inventive; and / or enjoys freedom to operate.

18. The computer-implemented method of any of claims 1 to 17, wherein the invention-related information is received as part of a training dataset for training one or more of the machine learning models, and wherein the invention-related information is used to train the one or more machine learning models.19 The computer-implemented method of any of claims 1 to 18, wherein the provided indication indicates an identity of the one or more sources that have the same DCO and one or more ECFs as the invention-related information, and optionally when dependent on claim 7, the provided indication indicates an identity of the one or more sources that have the same DCI and / or one or more EPCs as the invention-related information.

20. A computer-implemented method (300) of modifying data in a database, the method comprising:Receiving (302) one or more sources;Classifying (304) the one or more sources according to the computer- implemented method of any of claims 1 to 19;Generating (306) an indication of the DCO and the one or more ECFs for the source, and optionally when dependent on claim 7, the DCI and / or one or more EPCs for the source, wherein the indication represents data to be added to the database and / or data to be used to modify existing data in the database.

21. A computer- implemented method (400) of training a machine-learning model for implementing the computer-implemented method of any of claims 1 to 20, the method comprising:Receiving (402) one or more sources of a first dataset; andUsing (404) a first machine learning technique to annotate the one or more sources of the first dataset to indicate the DCO and the one or more ECFs for the source, and optionally when dependent on claim 7, the DCI and / or one or more EPCs for the source.

22. The computer-implemented method of claim 21, wherein an unsupervised learning technique is used to extract one or more linguistic key terms for the DCO and theone or more ECFs for the source, and optionally when dependent on claim 7, the DCI and / or one or more EPCs for the source.

23. The computer-implemented method (500) of any of claims 21 to 22, comprising:Receiving (502) one or more sources of a second dataset that is larger than the first dataset; andUsing (504) a second machine learning technique to train the one or more machine learning models from the one or more sources of the second dataset, wherein the annotated one or more sources of the first dataset are used to supervise the second machine learning technique.

24. The computer-implemented method of any of claims 1 to 23, wherein the source comprises one or more claims of a patent document.

25. The computer-implemented method of any of claims 1 to 24, in which selection of one or more sources is done at random.

26. The computer-implemented method of any of claims 1 to 24, in which the selection of the starting sources is determined by the user.

27. The computer-implemented method of any of claims 1 to 26, which also includes a third set of linguistic key terms for Essential Physical Conditions, which can either be entered by the user or obtained from the source, which can include any possible synonyms, hyponyms, coordinates, and hypernyms.

28. The computer-implemented method of any of claims 1 to 27, in which the linguistic key terms that are isolated from the sources are obtained from specifications of patents and patent claims, optionally the independent claims in each patent or patent application, and further optionally any examination report that is available for each patent or patent application.

29. The computer-implemented method of any of claims 1 to 28, in which the sources are obtained from patents and patent applications, which have been classified according to an official classification.

30. The computer-implemented method of any of claims 1 to 29, in which the sources include audio sources which have been converted into text.

31. An apparatus (1000) comprising a processor (1002) configured to perform the method of any of claims 1 to 30.

32. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any of claims 1 to 30.3133. A machine-readable medium (900) comprising instructions (902) which, when executed by a processor (904), cause the processor to carry out the method of any of claims 1 to 30.

Images