DETAILED ACTION
This Office Action is in response to Applicant's response to application filed on 26 March 2024. Currently, claims 1-20 are pending. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statements (IDS) submitted are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are clearly drawn to at least one of the four categories of patent eligible subject matter recited in 35 U.S.C. 101 (apparatus and method). Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claims 1 and 11 recite the abstract idea of intelligently determining and implementing distinct routines for entities by receiving entity data associated with an entity and generating at least one distinct routine for the entity as a function of the entity data and generating a functional model as a function of the at least one distinct routine and generating, a distinct name as a function of the entity data, wherein generating the distinct name comprises identifying a plurality of attributes of the entity data and determining a component word set as a function of the plurality of attributes and generating the distinct name as a function of the component word set. The claims are directed to a type of generating entity data. Under prong 1 of Step 2A, these claims are considered abstract because the claims are certain method of organizing human activity including commercial interactions (including business relations). Applicant’s claims are organizing human activity because the claims show receiving entity data (which can be considered business or human activity) and that entity data is organized by being operated on and new determinations generated from the entity data. Under prong 2 of Step 2A, the judicial exception is not integrated into a practical application because the claims (the judicial exception and any additional elements individually or in combination such as the apparatus comprising: at least a processor; a memory communicatively coupled to the at least a processor, the memory containing instructions configuring the at least a processor to perform steps and using at least a processor, generating, using the at least a processor, a user interface data structure comprising the at least one distinct routine, the functional model, and the distinct name and receiving, using a graphical user interface (GUI) communicatively connected to the at least a processor, the user interface data structure; and displaying, using the GUI, the user interface data structure) are not an improvement to a computer or a technology, the claims do not apply the judicial exception with a particular machine, the claims do not effect a transformation or reduction of a particular article to a different state or thing nor do the claims apply the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment such that the claims as a whole is more than a drafting effort designed to monopolize the exception. These limitations at best are merely implementing an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f). Under Step 2B, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements individually or in combination such as the apparatus comprising: at least a processor; a memory communicatively coupled to the at least a processor, the memory containing instructions configuring the at least a processor to perform steps and using at least a processor, generating, using the at least a processor, a user interface data structure comprising the at least one distinct routine, the functional model, and the distinct name and receiving, using a graphical user interface (GUI) communicatively connected to the at least a processor, the user interface data structure; and displaying, using the GUI, the user interface data structure (as evidenced by para [0011]-[0016] of applicant’s own specification) are well understood, routine and conventional in the field. Dependent claims 2-10, 12-20 also do not include additional elements that integrate the judicial exception into a practical application because the additional elements either individually or in combination are merely an extension of the abstract idea itself by further showing determine the at least one distinct routine based on a frequency of occurrence of a particular routine within the entity data and calculate a distance metric between each distinct routine of the at least one distinct routine and each institutional routine of at least one institutional routine; and determine the at least one distinct routine as a function of the distance metric and generate attribute training data, wherein the attribute training data comprises correlations between exemplary entity data and exemplary attributes; train an attribute classifier using the attribute training data; and identify the plurality of attributes using the trained attribute classifier and generate cohort training data, wherein the cohort training data comprises correlations between exemplary entity data and exemplary entity cohorts; train a cohort classifier using the cohort training data; and classify the entity data into one or more entity cohorts using the trained cohort classifier and determine the plurality of attributes as a function of the one or more entity cohorts and determine at least a candidate name as a function of the component word set; and determine the distinct name as a function of the at least a candidate name and generate component word combination training data, wherein the component word combination training data comprises correlations between exemplary component words and exemplary candidate names and generate intelligibility rating training data, wherein the intelligibility rating training data comprises correlations between exemplary component words and exemplary intelligibility ratings and generate appeal rating training data, wherein the appeal rating training data comprises correlations between exemplary component words and exemplary appeal ratings. Dependent claims 8-10, 18-20 do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements individually or in combination such as train a component word combination machine learning model using the component word combination training data; and determine the at least a candidate name using the trained component word combination machine learning model and train an intelligibility rating machine learning model using the intelligibility rating training data; and determine the distinct name using the trained intelligibility rating machine learning model and train an appeal rating machine learning model using the appeal rating training data; and determine the distinct name using the trained appeal rating machine learning model (as evidenced by para [0011]-[0016] of applicant’s own specification) are well understood, routine and conventional in the field.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 3-11, 13-20 are rejected under 35 U.S.C. 103 as being unpatentable over Raju (US 2021/0158181 A1) in view of Toplyn (US 2021/0103700 A1)
Claims 1 and 11:
Raju, as shown, discloses the following limitations:
An apparatus (and corresponding method) for intelligently determining and implementing distinct routines for entities, the apparatus comprising: at least a processor; a memory communicatively coupled to the at least a processor, the memory containing instructions configuring the at least a processor to (see para [0015]-[0017], showing equivalent computing functionality and components):
receive entity data associated with an entity (see para [0032], “The method 400 begins at block 405 where computing device 110 may obtain a benchmark profile dataset for the benchmark company and store it in memory 120. At block 410, computing device 110 may also obtain profile data sets for each of a plurality of candidate companies and store the plurality of profile data sets in memory 120. A company's profile data set may comprise the technographic, firmographic, and public information of that company. In some embodiments, computing device 110 may obtain technographic, firmographic, and public information from data providers 130 directly and/or by scanning the data providers 130 (e.g., scanning numerous web domains or mobile applications associated with data providers 130) periodically. Computing device 110 may utilize any appropriate data collection tactics (or combination thereof) to perform this scanning including web crawling, and natural language processing, for example. Because most technologies leave behind a footprint or “signature” that helps a web crawler identify it from other elements of a website or mobile app, computing device 110 may find and catalog these signatures across large numbers of web sites, and learn which companies are using a given technology. In addition, by utilizing web crawlers to monitor each site periodically, computing device 110 may observe when certain technologies appear or disappear. For technologies that leave no footprint (e.g., databases, CRMs and other technologies), computing device 110 may utilize natural language processing, which involves scanning and digesting unstructured data from numerous sources. For example, computing device 110 may scan and analyze text from job postings, social media, press releases and more to infer a relationship between a company and particular technology. In some embodiments, computing device 110 may also utilize customer surveys and crowdsourcing technologies to obtain this information. Computing device 110 may obtain firmographic and public information in similar ways. The technographic, firmographic, and public information of a company may be referred to as the company's profile data set.”);
generate at least one distinct routine for the entity as a function of the entity data (see para [0024]-[0025], “Computing device 110 may then utilize an appropriate ML algorithm to compare the profile dataset for each of the plurality of candidate companies to the benchmark dataset 120C in order to find candidate companies that are the most closely related to the benchmark company. The machine learning algorithm may be any appropriate ML algorithm such as the K-nearest neighbor algorithm, neighborhood component analysis algorithm, or the large margin nearest neighbors algorithm, for example. More specifically, for each of the plurality of candidate companies, computing device 110 may utilize the ML algorithm to determine a distance between the profile data set for that company and the benchmark profile data set. Computing device 110 may determine this distance for each dimension (e.g., CRM software used, company size, vertical, etc.) of the profile data set and benchmark profile data sets as well as an overall distance computation. The distance may be calculated as the Euclidean distance, cosine similarity, or any other suitable distance measurement between corresponding values of the profile data set for that company and the benchmark profile data set for each dimension as well as overall distance. For example, computing device 110 may determine the Euclidean distance between the employee count, revenue, market segment, and other dimensions of the profile data set for that candidate company and the benchmark profile data set.” where the different dimensions can be considered distinct routines given broadest reasonable interpretation);
generate a functional model as a function of the at least one distinct routine (see para [0024]-[0026], where the dimension data is being utilized with a ML algorithm);
generate a distinct name as a function of the entity data (see para [0038], where the candidate company can be considered a distinct name based on the benchmark company entity data), wherein generating the distinct name comprises: identifying a plurality of attributes of the entity data (Fig 4A, showing the candidate company is based on profile data sets from the benchmark entity data);
generate a user interface data structure comprising the at least one distinct routine, the functional model, and the distinct name (see para [0030], “ In some embodiments, a user may modify, via a graphical user interface of a server hosting the servers of generating profile graphs, the ML algorithm used to emphasize or give additional weight to certain dimensions of information. For example, module 120A may provide a user interface which allows the user to weigh certain dimensions such as network security software used, company size, and location for example more heavily than other dimensions. In addition, module 120A may learn which dimensions are more important after a certain number of attempts, and may thus learn which factors should be weighted more.”); and
a graphical user interface (GUI) communicatively connected to the at least a processor, the GUI configured to: receive the user interface data structure; and display the user interface data structure (see para [0030], “In some embodiments, a user may modify, via a graphical user interface of a server hosting the servers of generating profile graphs, the ML algorithm used to emphasize or give additional weight to certain dimensions of information. For example, module 120A may provide a user interface which allows the user to weigh certain dimensions such as network security software used, company size, and location for example more heavily than other dimensions. In addition, module 120A may learn which dimensions are more important after a certain number of attempts, and may thus learn which factors should be weighted more.”).
Raju, however, does not specifically disclose determining a component word set as a function of the plurality of attributes. In analogous art, Toplyn discloses the following limitations:
determining a component word set as a function of the plurality of attributes (see para [0127]-[0128], “To measure how likely it is that two particular words will share a common context it is possible, in an exemplary embodiment, to measure how close their word vectors are in an appropriate vector space. This vector space can be produced by a machine learning model trained on a very large corpus of audience-relevant text in step 432. An exemplary embodiment might use GOOGLE's pretrained WORD2VEC model trained on approximately 100 billion words from a GOOGLE NEWS dataset; that model includes word vectors for a vocabulary of 3 million English words and phrases. Other embodiments might use machine learning models that were also trained on data scraped from online social media; pop culture websites; online forums, message boards, and discussion groups; and other data sources that represent the anticipated knowledge, beliefs, and attitudes of the intended audience for the joke or jokes to be generated. Other embodiments might use machine learning models that are even more tailored to an individual user because they have also been trained on that user's most frequently visited websites, websites in the user's online search and browsing history, social media posts made by or “liked” by the user, the user's email or text messages, the user's communications with an AI virtual assistant, and other personal data sources. ] In implementations the first step in selecting two topic keywords from the list of candidates for topic keywords is step 434, which filters out candidates that are not in the vocabulary of the machine learning model being used. Decision step 436 determines whether fewer than two candidates for topic keywords remain after the filtration process. In implementations, if fewer than two candidates remain then the method exits, as shown at step 438, and the method will not attempt to generate a joke. In other implementations, if fewer than two candidates for topic keywords remain, then the method may attempt again to obtain input text. If two or more candidates for topic keywords remain then the similarity, based on the machine learning model being used, of each and every pair of remaining candidates is calculated in step 440. An exemplary embodiment might use the “similarity” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data to calculate the similarity of each pair of candidates. An exemplary embodiment would also filter out, in step 442, any pair of candidates with a calculated similarity less than zero; such pairs of candidates may tend to be too dissimilar to be useful as topic keywords.”); and
generating the distinct name as a function of the component word set (see para [0026], “Generating the one or more punch words may further include selecting a group of words stored in the one or more databases which are related to the one or more topic keywords, sorting the group of words into pairs, and calculating a wordplay score for each pair.” and see para [0376]-[0383])
It would have been obvious to one or ordinary skill in the art at the time of the invention to combine the teachings of Toplyn with Raju because component word sets based on a function of attributes enables more effective processing of language for context (see Toplyn, para [0002]-[0007]).
Moreover, it would have been obvious to one of ordinary skill in the art at the time of the invention to include the system for recognizing punchlines as taught by Toplyn in the system for identifying similar companies of Raju, since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
Claims 3 and 13:
Further, Raju discloses the following limitations:
calculate a distance metric between each distinct routine of the at least one distinct routine and each institutional routine of at least one institutional routine (see para [0013], “The present disclosure addresses the above-noted and other deficiencies by using a processing device to obtain a benchmark profile data set for a benchmark company and obtain a plurality of profile data sets, each of the plurality of profile data sets corresponding to a candidate company. Each of the plurality of profile data sets and the benchmark profile data set comprises firmographic, technographic, and public information. The processing device may utilize a machine learning algorithm to determine the distance between each of the plurality of profile data sets and the benchmark profile data set and build profile graphs indicating the distance. The processing device may determine one or more of the plurality of profile data sets that are most similar to the benchmark profile data set based on the determined distances; and identify the one or more candidate companies corresponding to the one or more profile data sets as companies most similar to the benchmark company. Although discussed with respect to determining similar companies, embodiments of the present disclosure may be used to determine similarities between data sets in various different applications (e.g., those mentioned above).”); and
determine the at least one distinct routine as a function of the distance metric (see para [0026]-[0031], showing distance between values of the profile graphs).
Claims 4 and 14:
Further, Raju discloses the following limitations:
generate attribute training data, wherein the attribute training data comprises correlations between exemplary entity data and exemplary attributes (see para [0013], “The present disclosure addresses the above-noted and other deficiencies by using a processing device to obtain a benchmark profile data set for a benchmark company and obtain a plurality of profile data sets, each of the plurality of profile data sets corresponding to a candidate company. Each of the plurality of profile data sets and the benchmark profile data set comprises firmographic, technographic, and public information. The processing device may utilize a machine learning algorithm to determine the distance between each of the plurality of profile data sets and the benchmark profile data set and build profile graphs indicating the distance. The processing device may determine one or more of the plurality of profile data sets that are most similar to the benchmark profile data set based on the determined distances; and identify the one or more candidate companies corresponding to the one or more profile data sets as companies most similar to the benchmark company. Although discussed with respect to determining similar companies, embodiments of the present disclosure may be used to determine similarities between data sets in various different applications (e.g., those mentioned above).”);
train an attribute classifier using the attribute training data (see para [0013], where using machine learning can be considered training); and
identify the plurality of attributes using the trained attribute classifier (see para [0026], “Based on the ML analysis, for each of the plurality of profile data sets, computing device 110 may create a profile graph showing the proximity of that profile data set to the benchmark profile data set in various dimensions. A profile graph can be implemented in any form of graphical representations with any kinds of graphics attributes such as, colors and shapes etc. FIGS. 2A-2C illustrate profile graphs 200A-200C respectively. Each of the profile graphs 200 illustrate the proximity of a profile data set 201 to the benchmark profile data set 202 in 4 dimensions A-D. Each dimension may represent a particular aspect of the candidate company and benchmark profile data sets. In the example of FIGS. 2A-2C, dimension A may correspond to employee count, dimension B may correspond to revenue, dimension C may correspond to market segment, and dimension D may correspond to CRM software used. The graphics attributes such as a length, thickness, color, and/or shape of a dimension may be utilized to represent certain characteristics of data, such as number of employees, an amount of revenue, etc. FIGS. 2A-2C illustrate four dimensions of Euclidean distance for ease of illustration, but as discussed herein, the profile graph for each of the plurality of profile data sets may include a Euclidean distance between corresponding values of the profile data set and benchmark profile data set for each dimension of information.” and see para [0013]).
Claims 5 and 15:
Further, Raju discloses the following limitations:
generate cohort training data, wherein the cohort training data comprises correlations between exemplary entity data and exemplary entity cohorts (see para [0013], “The present disclosure addresses the above-noted and other deficiencies by using a processing device to obtain a benchmark profile data set for a benchmark company and obtain a plurality of profile data sets, each of the plurality of profile data sets corresponding to a candidate company. Each of the plurality of profile data sets and the benchmark profile data set comprises firmographic, technographic, and public information. The processing device may utilize a machine learning algorithm to determine the distance between each of the plurality of profile data sets and the benchmark profile data set and build profile graphs indicating the distance. The processing device may determine one or more of the plurality of profile data sets that are most similar to the benchmark profile data set based on the determined distances; and identify the one or more candidate companies corresponding to the one or more profile data sets as companies most similar to the benchmark company. Although discussed with respect to determining similar companies, embodiments of the present disclosure may be used to determine similarities between data sets in various different applications (e.g., those mentioned above).”);
train a cohort classifier using the cohort training data (see para [0013], where using machine learning can be considered training); and
classify the entity data into one or more entity cohorts using the trained cohort classifier (see para [0013], where the most similar can be considered the classification given broadest reasonable interpretation).
Claims 6 and 16:
Further, Raju discloses the following limitations:
determine the plurality of attributes as a function of the one or more entity cohorts (see para [0026], “Based on the ML analysis, for each of the plurality of profile data sets, computing device 110 may create a profile graph showing the proximity of that profile data set to the benchmark profile data set in various dimensions. A profile graph can be implemented in any form of graphical representations with any kinds of graphics attributes such as, colors and shapes etc. FIGS. 2A-2C illustrate profile graphs 200A-200C respectively. Each of the profile graphs 200 illustrate the proximity of a profile data set 201 to the benchmark profile data set 202 in 4 dimensions A-D. Each dimension may represent a particular aspect of the candidate company and benchmark profile data sets. In the example of FIGS. 2A-2C, dimension A may correspond to employee count, dimension B may correspond to revenue, dimension C may correspond to market segment, and dimension D may correspond to CRM software used. The graphics attributes such as a length, thickness, color, and/or shape of a dimension may be utilized to represent certain characteristics of data, such as number of employees, an amount of revenue, etc. FIGS. 2A-2C illustrate four dimensions of Euclidean distance for ease of illustration, but as discussed herein, the profile graph for each of the plurality of profile data sets may include a Euclidean distance between corresponding values of the profile data set and benchmark profile data set for each dimension of information.”)
Claims 7-10, 17-20:
Raju does not specifically disclose determine at least a candidate name as a function of the component word set. In analogous art, Toplyn discloses the following limitations:
determine at least a candidate name as a function of the component word set (see para [0127]-[0128], “To measure how likely it is that two particular words will share a common context it is possible, in an exemplary embodiment, to measure how close their word vectors are in an appropriate vector space. This vector space can be produced by a machine learning model trained on a very large corpus of audience-relevant text in step 432. An exemplary embodiment might use GOOGLE's pretrained WORD2VEC model trained on approximately 100 billion words from a GOOGLE NEWS dataset; that model includes word vectors for a vocabulary of 3 million English words and phrases. Other embodiments might use machine learning models that were also trained on data scraped from online social media; pop culture websites; online forums, message boards, and discussion groups; and other data sources that represent the anticipated knowledge, beliefs, and attitudes of the intended audience for the joke or jokes to be generated. Other embodiments might use machine learning models that are even more tailored to an individual user because they have also been trained on that user's most frequently visited websites, websites in the user's online search and browsing history, social media posts made by or “liked” by the user, the user's email or text messages, the user's communications with an AI virtual assistant, and other personal data sources. ] In implementations the first step in selecting two topic keywords from the list of candidates for topic keywords is step 434, which filters out candidates that are not in the vocabulary of the machine learning model being used. Decision step 436 determines whether fewer than two candidates for topic keywords remain after the filtration process. In implementations, if fewer than two candidates remain then the method exits, as shown at step 438, and the method will not attempt to generate a joke. In other implementations, if fewer than two candidates for topic keywords remain, then the method may attempt again to obtain input text. If two or more candidates for topic keywords remain then the similarity, based on the machine learning model being used, of each and every pair of remaining candidates is calculated in step 440. An exemplary embodiment might use the “similarity” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data to calculate the similarity of each pair of candidates. An exemplary embodiment would also filter out, in step 442, any pair of candidates with a calculated similarity less than zero; such pairs of candidates may tend to be too dissimilar to be useful as topic keywords.”); and
determine the distinct name as a function of the at least a candidate name (see para [0026], “Generating the one or more punch words may further include selecting a group of words stored in the one or more databases which are related to the one or more topic keywords, sorting the group of words into pairs, and calculating a wordplay score for each pair.” and see para [0376]-[0383]).
generate component word combination training data, wherein the component word combination training data comprises correlations between exemplary component words and exemplary candidate names (see para [0131], “In implementations, after two topic keywords have been selected as in FIG. 4, a joke of Type #1 is generated from them. FIG. 5 shows an exemplary embodiment of a method 500 of generating a first type of joke (Type #1). Block 510 represents the two topic keywords that were outputted in step 446. In step 512, the machine learning model trained in step 432, or another language model, is used to list the words that are most similar to each topic keyword. That is, for each of the two topic keywords, a list is created of the top fifty, for example (which number in implementations may be selected by the admin), words that occur most frequently in contexts where the topic keyword appears. An exemplary embodiment might use the “most_similar” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data to build the list of most similar words, which we'll refer to as related words. The topic keyword or one or more of its related words may consist of more than one token. The list of related words for a topic keyword may be filtered to exclude words that are poorly suited for use in assembling punch words, by non-limiting example, duplicate words, misspelled words, acronyms, abbreviations, words with non-alphabetic characters, or words unrelated to the sense of the topic keyword as it is used in the topic sentence. In other implementations, related words, associated words, or most similar words for a topic keyword, including words that occur most frequently in contexts where the topic keyword appears, may also be generated by a language model such as, by non-limiting example, OPENAI's GPT-3 or may be selected from a database or databases such as, by non-limiting example, a spreadsheet, knowledge base, graph database, knowledge graph, relational database, semantic network, entity relationship model, dictionary, thesaurus, ontology, lexicon, or WORDNET (created by George A. Miller), or may be selected by using a common knowledge model such as those described below in step 2012 or by using any other language model or database that allows a determination to be made of how closely connected two words, two word chunks, or a word and a word chunk are in the minds of most people in the intended audience for the joke. In still other implementations, related words may be selected from a database or databases including words related to an entity or entities that are not a topic keyword, by non-limiting example, words related to dogs, which related words may be used to generate or recognize jokes from the perspective of a dog.” and see para [0127]-[0128], [0164]-[0165], [0252]-[0254]);
train a component word combination machine learning model using the component word combination training data (see para [0127]-[0128], “To measure how likely it is that two particular words will share a common context it is possible, in an exemplary embodiment, to measure how close their word vectors are in an appropriate vector space. This vector space can be produced by a machine learning model trained on a very large corpus of audience-relevant text in step 432. An exemplary embodiment might use GOOGLE's pretrained WORD2VEC model trained on approximately 100 billion words from a GOOGLE NEWS dataset; that model includes word vectors for a vocabulary of 3 million English words and phrases. Other embodiments might use machine learning models that were also trained on data scraped from online social media; pop culture websites; online forums, message boards, and discussion groups; and other data sources that represent the anticipated knowledge, beliefs, and attitudes of the intended audience for the joke or jokes to be generated. Other embodiments might use machine learning models that are even more tailored to an individual user because they have also been trained on that user's most frequently visited websites, websites in the user's online search and browsing history, social media posts made by or “liked” by the user, the user's email or text messages, the user's communications with an AI virtual assistant, and other personal data sources. ] In implementations the first step in selecting two topic keywords from the list of candidates for topic keywords is step 434, which filters out candidates that are not in the vocabulary of the machine learning model being used. Decision step 436 determines whether fewer than two candidates for topic keywords remain after the filtration process. In implementations, if fewer than two candidates remain then the method exits, as shown at step 438, and the method will not attempt to generate a joke. In other implementations, if fewer than two candidates for topic keywords remain, then the method may attempt again to obtain input text. If two or more candidates for topic keywords remain then the similarity, based on the machine learning model being used, of each and every pair of remaining candidates is calculated in step 440. An exemplary embodiment might use the “similarity” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data to calculate the similarity of each pair of candidates. An exemplary embodiment would also filter out, in step 442, any pair of candidates with a calculated similarity less than zero; such pairs of candidates may tend to be too dissimilar to be useful as topic keywords.” and see para [0131], [0164]-[0165], [0252]-[0254]); and
determine the at least a candidate name using the trained component word combination machine learning model (see para [0127]-[0128], “To measure how likely it is that two particular words will share a common context it is possible, in an exemplary embodiment, to measure how close their word vectors are in an appropriate vector space. This vector space can be produced by a machine learning model trained on a very large corpus of audience-relevant text in step 432. An exemplary embodiment might use GOOGLE's pretrained WORD2VEC model trained on approximately 100 billion words from a GOOGLE NEWS dataset; that model includes word vectors for a vocabulary of 3 million English words and phrases. Other embodiments might use machine learning models that were also trained on data scraped from online social media; pop culture websites; online forums, message boards, and discussion groups; and other data sources that represent the anticipated knowledge, beliefs, and attitudes of the intended audience for the joke or jokes to be generated. Other embodiments might use machine learning models that are even more tailored to an individual user because they have also been trained on that user's most frequently visited websites, websites in the user's online search and browsing history, social media posts made by or “liked” by the user, the user's email or text messages, the user's communications with an AI virtual assistant, and other personal data sources. ] In implementations the first step in selecting two topic keywords from the list of candidates for topic keywords is step 434, which filters out candidates that are not in the vocabulary of the machine learning model being used. Decision step 436 determines whether fewer than two candidates for topic keywords remain after the filtration process. In implementations, if fewer than two candidates remain then the method exits, as shown at step 438, and the method will not attempt to generate a joke. In other implementations, if fewer than two candidates for topic keywords remain, then the method may attempt again to obtain input text. If two or more candidates for topic keywords remain then the similarity, based on the machine learning model being used, of each and every pair of remaining candidates is calculated in step 440. An exemplary embodiment might use the “similarity” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data to calculate the similarity of each pair of candidates. An exemplary embodiment would also filter out, in step 442, any pair of candidates with a calculated similarity less than zero; such pairs of candidates may tend to be too dissimilar to be useful as topic keywords.” and see para [0131], [0164]-[0165], [0252]-[0254])
generate intelligibility rating training data, wherein the intelligibility rating training data comprises correlations between exemplary component words and exemplary intelligibility ratings (see para [0292], “At decision step 1718 the method uses natural language processing tools to pair punch word A with each topic keyword candidate in turn. The method then calculates a relatedness score for every pairing of punch word A and each of the topic keyword candidates and determines whether any pairing of punch word A and a topic keyword candidate has a relatedness score which is better than a certain preset threshold, which threshold may be determined by the admin in implementations and may be set to any desired value. As described in steps 432 through 440 inclusive, an exemplary embodiment may calculate a relatedness score by measuring how close the word vectors for punch word A and a topic keyword candidate are in an appropriate vector space by using, as a non-limiting example, the “similarity” function of GOOGLE's pretrained WORD2VEC model trained on GOOGLE NEWS data. Other embodiments of this system may calculate a relatedness score using other language models trained on other data corpuses, as described herein and in the drawings with respect to step 432. Still other embodiments of this system may calculate a relatedness score using any of the methods described herein and in the drawings with respect to step 512.” and see para [0314], “Clarity score: The method calculates the relatedness score between topic keyword A and punch word A and the relatedness score between topic keyword B and punch word B and uses both of those scores to get the clarity score; in an exemplary embodiment, an average of the two relatedness scores may be taken. The relatedness scores may be calculated as described in decision step 1718. The more closely related the punch words are to the topic keywords, the more understandable the joke, the better the clarity score, and the better the funniness score. This process is related to step 512, in which related words are listed that are most similar to each topic keyword.”);
train an intelligibility rating machine learning model using the intelligibility rating training data (see para [0314], “Clarity score: The method calculates the relatedness score between topic keyword A and punch word A and the relatedness score between topic keyword B and punch word B and uses both of those scores to get the clarity score; in an exemplary embodiment, an average of the two relatedness scores may be taken. The relatedness scores may be calculated as described in decision step 1718. The more closely related the punch words are to the topic keywords, the more understandable the joke, the better the clarity score, and the better the funniness score. This process is related to step 512, in which related words are listed that are most similar to each topic keyword.” and see para [0315] and Fig 5); and
determine the distinct name using the trained intelligibility rating machine learning model (see para [0314], “Clarity score: The method calculates the relatedness score between topic keyword A and punch word A and the relatedness score between topic keyword B and punch word B and uses both of those scores to get the clarity score; in an exemplary embodiment, an average of the two relatedness scores may be taken. The relatedness scores may be calculated as described in decision step 1718. The more closely related the punch words are to the topic keywords, the more understandable the joke, the better the clarity score, and the better the funniness score. This process is related to step 512, in which related words are listed that are most similar to each topic keyword.” and see para [0315] and Fig 5)
generate appeal rating training data, wherein the appeal rating training data comprises correlations between exemplary component words and exemplary appeal ratings; train an appeal rating machine learning model using the appeal rating training data; and determine the distinct name using the trained appeal rating machine learning model (see para [0312]-[0315], “Interest score: The method calculates the degree to which topic keyword A and topic keyword B are related to each other as measured by their relatedness score, which may be calculated as described in decision step 1718. The less related that topic keyword A and topic keyword B are to each other, the more attention-getting they may be when appearing together in the topic, the better the interest score, and the better the funniness score. This process is related to step 440 through step 444 inclusive, in which two topic keywords are selected in part by calculating how dissimilar they are. Position score: The method uses natural language processing tools to measure how close punch word A and punch word B are to the end of the response. The closer to the end, the better the position score and the better the funniness score. This process is related to that described above in relation to FIG. 2, in which jokes are assembled by placing the punch words at the end. Clarity score: The method calculates the relatedness score between topic keyword A and punch word A and the relatedness score between topic keyword B and punch word B and uses both of those scores to get the clarity score; in an exemplary embodiment, an average of the two relatedness scores may be taken. The relatedness scores may be calculated as described in decision step 1718. The more closely related the punch words are to the topic keywords, the more understandable the joke, the better the clarity score, and the better the funniness score. This process is related to step 512, in which related words are listed that are most similar to each topic keyword. Wordplay score: The method calculates the wordplay score between punch word A and punch word B using the process shown in FIG. 6. The better the wordplay, the better the wordplay score and the better the funniness score. This process is related to step 516 and step 520, in which punch words are selected using their wordplay score.” and Fig 5)
It would have been obvious to one of ordinary skill in the art at the time of the invention to include the system for recognizing punchlines as taught by Toplyn in the system for identifying similar companies of Raju, since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
Claims 2 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Raju and Toplyn, as applied above, and further in view of Roache et al. (US 8306923 B1)
Claim 2:
Raju and Toplyn do not specifically disclose determining the at least one distinct routine based on a frequency of occurrence of a particular routine within the entity data. In analogous art, Roache discloses the following limitations:
determine the at least one distinct routine based on a frequency of occurrence of a particular routine within the entity data (col 10, line 36-60, “ For example, a trend showing the increasing volume of parcel delivery transactions (e.g., executed by a particular carrier) may correspond to a rating indicating a greater legitimacy of the business entity. Thus, referring to FIG. 5, Business Entity A, which has been involved in an increasing number of parcel delivery transactions over the past 5 years, may be rated higher than Business Entity B, which has had a constant number of parcel delivery transactions. Furthermore, both A and B may be rated higher than Business Entity C, which has had a steadily decreasing number of parcel delivery transactions over the same amount of time, perhaps indicating a decline in Business Entity C's business. Similarly, a trend showing the increasing frequency of parcel delivery transactions may correspond to a rating indicating a greater legitimacy of the business entity. Thus, a business entity that has gone from engaging a particular common carrier to deliver parcels, for example, twice a week to engaging the common carrier to deliver parcels four times a week may be rated higher than another business entity that has slipped from using the common carrier twice a week to using the common carrier only once a week. In addition to volume and frequency, trends in the value of the common carrier's services used may also be taken into account when determining the rating.” and col 12, line 32-50, “In other embodiments, a method for certifying a business entity is provided. Referring to Block 100 of FIG. 6, input is received regarding a location of the business entity visited as part of a parcel delivery transaction. As described above, the input may be received automatically, for example as a result of obtaining the business entity's signature upon receiving a parcel (e.g., via a handheld device, such as a DIAD), or manually, such as when an employee of the common carrier (e.g., the driver making the delivery) provides input indicating that the delivery was successfully completed at the specified address or confirming that the business entity is located at a particular address in the process of conducting other parcel delivery activities, as described above. In Block 102, the business entity is then associated with a verified address corresponding to the location visited. In this way, the verified address corresponds to a physical location at which the business entity is actually doing business in some capacity and is able to receive parcels.”)
It would have been obvious to one or ordinary skill in the art at the time of the invention to combine the teachings of Roache with Raju and Toplyn because including a frequency of occurrence of a particular routine within the entity data can be helpful in reducing fraudulent dealings (see Roache, col 1, line 15-50).
Moreover, it would have been obvious to one of ordinary skill in the art at the time of the invention to include the system for certifying business entities as taught by Roache in the Raju and Toplyn combination, since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Bennett et al. (US 2021/0019357 A1), a system for transmitting questionnaires wherein the questionnaires include a set of questions that relate to an internal entity experience attribute and/or an external entity experience attribute. The device may receive a set of questionnaire responses and correlate the set of questionnaire responses with contextual data relating to an entity and using a machine learning model trained based on data relating to one or more other entities, the set of questionnaire responses and the contextual data to identify a set of features and a set of feature ranks
Ramasamy et al. (AU 2019219741 A1), a system generating an architecture diagram, the system including an input processor configured to receive, from a terminal, entity data associated with a plurality of entities of an architecture and path data associated with a plurality of paths that correspond to interconnections between the plurality of entities, a machine learning processor that utilizes a training dataset to assess whether the entities defined by the entity data are correctly interconnected as defined by the path data and to predict an entity and/or path suitable to add to the architecture, an advice generator that receives an assessment from the machine learning processor, prepares a recommendation based on the assessment to include the predicted entity and/or path, and communicates the recommendation to the terminal, wherein the assessment received from the machine learning processor indicates a proposed change to either insert a new entity between entities of the architecture or to replace an existing entity of the architecture with the new entity, wherein the recommendation includes the proposed change and the advice generator is further configured to receive an indication of acceptance or rejection of the recommendation from the terminal, and when the recommendation is rejected, the advice generator is configured to communicate an indication to the machine learning processor that the proposed change in the recommendation was rejected, wherein in response to the indication that the proposed change was rejected, the machine learning processor is further configured to update information in training the dataset to reflect that the proposed change was rejected
Awan “What is Named Entity Recognition (NER)? Methods, Use Cases, and Challenges”, a blog on Named Entity Recognition (NER) which is a sub-task of information extraction in Natural Language Processing (NLP) that classifies named entities into predefined categories such as person names, organizations, locations, medical codes, time expressions, quantities, monetary values, and more.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SUJAY KONERU whose telephone number is 571-270-3409. The examiner can normally be reached on Monday-Friday, 9 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Patricia Munson can be reached on 571- 270-5396. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SUJAY KONERU/
Primary Examiner, Art Unit 3624
Prosecution Insights