Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
2. Applicant’s arguments with respect to claim 1 have been considered but are not persuasive. Specifically, the newly added limitation: “apply Latent Dirichlet Allocation (LDA) topic modeling algorithm to the extracted words to extract a plurality of topics” is taught by Mehta in Section 4.1, which discloses that open vocabulary personality detection methods extract language features from text including clusters of semantically related words and that Latent Dirichlet Allocation (LDA) is used for forming clusters of semantically related words.
Applicant’s arguments with respect to claim 1 have been considered but are moot because the new ground of rejection relies on newly cited McCourt to teach newly added limitation: “…output the extracted plurality of topics through a display device” and newly cited Preuss to teach “receive, from an administrator, selection, information indicating which of the plurality of topics are associated with the personality, and exclude a word included in a topic that is not indicated by the selection information.”
Therefore, the rejection of claims 1-6 and 8-10 is maintained
3. Claims 1-6 and 8-10 are rejected under 35 U.S.C. 103 as being unpatentable over MacKay (US 2022/0245354) in view of in view of Mehta et al “Recent Trends in deep learning based personality detection” herein Mehta, further in view of McCourt (US 2022/0383863), further in view of Preuss (US 2021/0233030) and further in view of Leonardi et al. “Multilingual Transformer-Based Personality Traits Estimation” herein Leonardi.
Regarding Claim 1:
MacKay discloses a personality assessment device comprising (Figs. 2 and 3):
at least one processor comprising processing circuitry (MacKay: p[0057]-[0059] discloses processors)
and a memory comprising one or more storage media storing instructions, which when executed individually or collectively by the at least one processor (MacKay: p[0057]-[0059] discloses memory), causes the device to:
transmit, to a terminal, a message that includes a plurality of interview questions and at least one directive sentence and to receive, from the terminal, a first response to each of the plurality of interview questions and a second response to the at least one directive sentence (MacKay: p[0006] creates a personality quiz including a plurality of questions; p[0036] discloses an instruction based question format i.e., directive; p[0035] discloses that the system has an unending number of questions for the user to answer, the user is essentially directed to interact with these prompts);
automatically tokenize the first response and the second response into words using natural language processing (MacKay: p[0056] tokenizes words using NLP and pre-processing steps), extract words having predetermined parts of speech including adjectives, verbs, and nouns, (MacKay: p[0052], p[0086] and p[0010] altogether disclose the extraction of parts of speech including verbs, nouns, adjectives and adverbs) and select at least some of the extracted words based on a computed correlation with personality traits using a trained correlation model stored in the memory (MacKay: p[0161] explicitly mentions selecting individual words, these words are matched to emotion/cognition/sentiment-based categories (i.e., this data can be used to infer personality traits) this is done by the emotion-cognitive sensor; p[0067] discloses that machine learning models can be trained on the rules themselves, the trained models can locate similar co-occurrence vectors, and their output directly effects the emotion-cognitive sensor additionally this emotion-cognitive sensor can update/generate new rules)
;
MacKay does not explicitly disclose:
apply Latent Dirichlet Allocation (LDA) topic modeling algorithm to the extracted words to extract a plurality of topics.
However, Mehta discloses apply Latent Dirichlet Allocation (LDA) topic modeling algorithm to the extracted words to extract a plurality of topics (Mehta: teaches personality detection from language use, extracting a collection of language features from text and producing topic clusters of semantically related words using Latent Dirichlet Allocation (LDA). The result of the LDA step is multiple topic clusters, each cluster being of semantically related words for personality detection).
MacKay and Mehta are combinable because they are both from the same field of endeavor, deep learning based personality and emotion-cognition detection, e.g., both disclose systems for using multi-layered neural networks for detecting the emotional state and responses of users. It would have been obvious to one of ordinary skill in the art to incorporate LDA-based topic modeling of Mehta into the emotion-cognitive natural language processing of MacKay to enhance personality prediction accuracy by extracting semantically meaningful groups of words (topics) and selecting only those with demonstrated personality relevance. The motivation for doing so would be “Reliable correlations of writing style (e.g., frequency of word use) with personality were found by some of the earliest works” as disclosed in Section 4.1 of Mehta and LDA is a method of finding reliable correlations between words and writing styles to detect personality.
MacKay and Mehta do not explicitly disclose output the extracted plurality of topics through a display device, receive, from an administrator, selection, information indicating which of the plurality of topics are associated with the personality, and exclude a word included in a topic that is not indicated by the selection information;
However, McCourt discloses:
output the extracted plurality of topics through a display device (McCourt: ¶[0094]-[0096] discloses receiving user/administrator input that names/labels topics), …, and exclude a word included in a topic that is not indicated by the selection information receive, from an administrator, selection, information indicating which of the plurality of topics are associated with the personality (McCourt: ¶[0120]-[0121] discloses that topic identifiers are removed/updated and removes words corresponding to a topic when updating the topic model output);
MacKay and Mehta in view of McCourt are combinable because they are all from pertinent fields of endeavor. MacKay and Mehta disclose systems for using multi-layered neural networks for detecting the emotional state and responses of users, and McCourt discloses a topic-display and topic-output updating/removal technique. It would have been obvious to one of ordinary skill in the art to incorporate this feature into MacKay and Mehta’s personality trait/topic clustering techniques. The motivation for doing so is “Topic modeling in written and verbal communications can be extremely useful for grouping a large number of communications for review, analysis, or intervention.” as disclosed in ¶[0004] of McCourt.
MacKay, Mehta and McCourt do not explicitly disclose receive, from an administrator, selection, information indicating which of the plurality of topics are associated with the personality.
However, Preuss discloses receive, from an administrator, selection, information indicating which of the plurality of topics are associated with the personality (Preuss: ¶[0054] discloses an interviewer (administrator) may select questions to customize the personality traits they are looking for).
MacKay, Mehta and McCourt in view of Preuss are combinable because they all come from pertinent fields of endeavor, MacKay and Mehta disclose methods for disclose systems for using multi-layered neural networks for detecting the emotional state and responses of users, and McCourt discloses a topic-display and topic-output updating/removal technique. McCourt merely does not disclose that the topics chosen are related to personality. Preuss discloses identifying associated personality aspect by applying a natural language classifier to detect words and phrases. McKay and Mehta provide the substantive personality-related analytics, McCourt provides topic extraction/display and exclusion of updated settings and Preuss provides the missing administrator selection information being based on personality traits. It would have been obvious to one of ordinary skill in the art to incorporate this system. The motivation for doing so is “The customized data structure of the interview question mappings improves processing efficiency of the system candidate score calculations” as disclosed in ¶[0036] of Preuss.
MacKay, Mehta, McCourt and Preuss do not explicitly disclose and input the selected words into a transformer- based language model trained to output numerical scores for personality traits, and to predict personality of a user of the terminal based on outputs from the model, however, Leonardi discloses this limitation (Leonardi: Figure 4 discloses inputs of tokenized words into a transformer (BERT-based) encoder; Introduction the neural network “exploits” sentences embeddings to perform a regression on the objective personality trait which provides numerical output and scores; section 5.2 discloses regression predicts on of the five personality traits).
The combination of Mackay, Mehta, McCourt and Preuss in view of Leonardi are combinable because they are from the same field of endeavor of emotion and personality classification, e.g., both disclose system that perform natural language processing on textual input to identify emotional and cognitive information. It would have been obvious to one of ordinary skill in[ the art before the effective filing date of the claimed invention to disclose inputting tokenized answers to personality questions to a transformer based model. The motivation for doing so is “the wider adoption of deep learning techniques and the incremental availability of data and machine resources, novel studies emerged” as disclosed by Leonardi in Section 2.4. Additionally, MacKay states in paragraph [0068] that the machine learning models engine 320 can include any machine learning model or models from transformer models, such as BERT, RoBERTa etc.
Regarding Claim 2:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further discloses the personality assessment device of claim 1, wherein each of the plurality of interview questions is a question for predicting the personality of the user, requiring a subjective response (MacKay: p[0249] a profile is generated with personality traits, p[0111] says this is predicted from input which can include user data or transcribed interview data/surveys, p[0004]-[0005] describes subjective data), and the at least one directive sentence includes a directive sentence that requests a response to a plurality of adjectives that are thought to describe the user well (MacKay: p[0251] discloses the system solicits descriptors of personality, which often come inf the form of adjectives (e.g., honest, thoughtful, team-player etc.) to point to an ideal candidate, the word personality traits is reasonably broad enough to include adjectives considering they are a core component of all language and personality prediction) and a directive sentence that requests a response to a strength and a weakness of the personality of the user (MacKay: p[0095] discloses the system handles trait-related descriptors (ego, curiosity) and associates them with linguistic patterns and their emotional effects that can be deemed positive or negative such as self-esteem, empathy or patience, p[0103] discloses the user data can determine user ego that is positive negative and neutral based on their answers, responses and input (i.e., strength or weakness as it evaluates the user responses for their self-perception)).
Regarding Claim 3:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further discloses the personality assessment device of claim 1, wherein the device is further configured to transcribe, using an automatic speech recognition model executed by the at least one processor, at least one of the first response in the form of speech data and the second response in the form of speech data to the form of text data (MacKay: Fig. 8 S802 discloses receiving textual and or voice input, p[0074] discloses transcribed human speech input).
Regarding Claim 4:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further discloses the personality assessment device of claim 1, wherein the device is further configured to perform part-of-speech tagging on the tokenized responses and extract words corresponding to at least one of an adjective, a verb, and a noun from among words included in the responses, and to select a subset of the extracted words based on correlation with personality traits using the trained correlation model (MacKay: p[0052], p[0086] and p[0010] disclose the extracting parts of speech from user input including verbs, nouns, adjectives and adverbs recognized in the users response, these important parts of speech are grouped into emotions or cognition, each specific part of speech rule has its own category, these rule categories can include personality traits and emotional states).
Regarding Claim 5:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further disclose the personality assessment device of claim 4, except wherein the preprocessing unit is configured to extract a plurality of topics by applying a Latent Dirichlet Allocation (LDA) method to the extracted words and to select some words from among the extracted words based on correlation between each of the plurality of topics and the personality (Mehta: Section 4.1 paragraphs 11-12 discloses applying LDA to text responses to identify latent topics and using correlation between topics and personality traits to select important words/features for downstream personality classification).
It would have been obvious to one of ordinary skill in the art to incorporate LDA-based topic modeling of Mehta into the emotion-cognitive natural language processing of MacKay to enhance personality prediction accuracy by extracting semantically meaningful groups of words (topics) and selecting only those with demonstrated personality relevance. The motivation for doing so would be to use LDA as a complementary technique for categorization accuracy.
Regarding Claim 6:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further disclose the personality assessment device of claim 5, wherein the device is further configured to compute the correlation between each of the plurality of topics and the personality using a stored word list that includes a plurality of weighted words and to exclude a word included in a topic having the correlation of a predetermined threshold or less among the plurality of topics (Mehta: Section 4.1 paragraph 10, and 11 discloses entropy-based weights assigned to features extracted from text (which satisfies weighted words), LDA topics (this is mentioned explicitly), weights reflecting trait relevance (correlation with personality) and it is understood that filtering features with lower entropy/informative value (i.e., discarding low-signal topics or features) is analogous to excluding a word from a topic having correlation below a threshold).
It would have been obvious to one of ordinary skill in the art to incorporate LDA-based topic modeling of Mehta into the emotion-cognitive natural language processing of MacKay to configure the preprocessing unit of Mackay’s personality assessment system to exclude words in topics having correlation below a predetermined threshold as taught by Mehta. The motivation for doing so would be to reflect trait relevance and filter out features with lower informative value.
Regarding Claim 8:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further disclose the personality assessment device of claim 5, wherein the preprocessing unit is configured to compute the correlation between each of the plurality of topics and the personality using a correlation decision model that outputs correlation with a topic and to exclude a word included in the topic having the correlation of the predetermined threshold or less among the plurality of topics (Mehta: Section 4.1 discloses ML-kNN classifier weighted with entropy theory to assess feature correlation and use of learned model outputs to determine topic-personality relevance).
It would have been obvious to one of ordinary skill in the art to incorporate LDA-based topic modeling of Mehta into the emotion-cognitive natural language processing of MacKay, to incorporate the ML-kNN classifier weighted with entropy theory. The motivation for doing so is that it enables the system to exclude topics that fall below relevance threshold, thereby streamlining the prediction efficiency.
Regarding Claim 9:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further discloses the personality assessment device of claim 1, wherein the device is further configured to embed the selected words into a numerical format (MacKay: p[0064] discloses the use of vector embeddings meaning the user input is being transformed into numerical data)and input the embedding into a trained personality prediction model implemented using a transformer-based language model (Leonardi: Figure 4 discloses inputs of tokenized words into a transformer (BERT-based) encoder; Introduction the neural network “exploits” sentences embeddings (numerical format) to perform a regression on the objective personality trait which provides numerical output and scores).
Regarding Claim 10:
The combination of MacKay, Mehta, McCourt, Preuss and Leonardi further discloses the personality assessment device of claim 9, wherein the personality prediction model is a transformer-based pretrained language model stored in memory (Leonardi discloses using a BERT-based transformer model in Fig. 4 that is pretrained), and the personality prediction model is configured to output prediction points for top five adaptive factors and top five maladaptive personality factors (MacKay: p[0095] discloses the system handles trait-related descriptors (ego, curiosity) it associates these with emotional effects that can be deemed positive or negative such as self-esteem, empathy or patience, p[0103] discloses the user data can determine user ego that is positive negative and neutral based on their answers, responses and input (i.e., maladaptive and adaptive traits are found and output as it evaluates the user responses for their self-perception).
The combination of Mackay, Mehta, McCourt and Preuss in view of Leonardi are combinable because they are from the same field of endeavor of emotion and personality classification, e.g., both disclose system that perform natural language processing on textual input to identify emotional and cognitive information. It would have been obvious to one of ordinary skill in[ the art before the effective filing date of the claimed invention to disclose inputting tokenized answers to personality questions to a transformer based model. The motivation for doing so is “the wider adoption of deep learning techniques and the incremental availability of data and machine resources, novel studies emerged” as disclosed by Leonardi in Section 2.4. Additionally, MacKay states in paragraph [0068] that the machine learning models engine 320 can include any machine learning model or models from transformer models, such as BERT, RoBERTa etc.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IAN SCOTT MCLEAN whose telephone number is (703)756-4599. The examiner can normally be reached "Monday - Friday 8:00-5:00 EST, off Every 2nd Friday".
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Hai Phan can be reached at (571) 272-6338. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format.
For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/IAN SCOTT MCLEAN/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654