Research and Development

Artificial Intelligence and the Contextual Recommendation of Advice and Wisdom


Wise people share advice and wisdom in the forms of allegories, anecdotes, aphorisms, apologues, fables, folklore, historical analogues, jokes, literature, lyrics, parables, poems, proverbs, quotations, stories, and witticisms. It is a multidisciplinary challenge to build artificial intelligence systems capable of these tasks.

Towards solving this challenge, a new approach is presented: story-based search and recommendation. In this approach, individuals provide stories to retrieve content that is to be useful for selected story characters. The stories they provide could be real-world stories and the characters they select, in these cases, could be themselves or other people. Interestingly, individuals’ social media posts and feeds could be of similar use for establishing contexts for search and recommendation.

While the comprehension of story and social situations are key to the contextual search for and recommendation of advice and wisdom, there is a need for overarching architectures, frameworks, and models for artificial intelligence systems to best do so at scale. Intelligent coaching systems are indicated to be of use in these regards.

Search and recommender system approaches are considered here, in addition to dialogue systems and chatbots, because intelligent coaching systems, at scale, are envisioned as extensively reusing items, e.g., messages of advice, rather than utilizing natural-language generation algorithms to contextually produce new such messages for individuals in an on-the-fly manner.

Applications of the technologies under discussion include social media, education, library and information science, knowledge management, and history.

Story-based Search and Recommendation

In story-based search and recommendation, individuals provide stories to retrieve content that is to be useful for selected story characters. These provided stories can be fictional or real-world stories.

Use case scenarios for fictional stories include their uses in training, testing, and evaluation. These datasets could utilize metadata for indicating stories’ reading levels and other developmental narratological factors.

Use case scenarios for real-world stories include those where individuals seek to retrieve content for themselves and those where individuals, e.g., peers, teachers, or guidance counselors, seek to retrieve content for other individuals or audiences.

Stories provide a natural means of establishing cognitive contexts. Viewing them in this way, active or conversational story comprehension can be considered. Narratees can ask questions of narrators during conversational processes of narration. Vague or partial cognitive story comprehension contexts can inform narratees’ processes of forming questions for narrators about unfolding narratives.

In incremental story-based search and recommendation, individuals engage in dialogues, narrating to artificial intelligence systems, and receive dynamically updating lists of recommended content for selected story characters. These recommended items could include question items and individuals could select these to have systems ask them them in unfolding dialogues. As individuals narrate to and answer questions from artificial intelligence systems incrementally comprehending their stories, content recommendations for selected characters would be provided.

Story and Social Comprehension

In order for artificial intelligence systems to be able to search for and recommend content for selected characters in provided stories, these systems should be able to comprehend stories.

One thing which separates machine reading comprehension from text processing is inferencing. Taxonomies describing reading-related inferences make distinctions between: automatic and strategic; online and offline; text-connecting, knowledge-based, and extratextual; local and global; coherence and elaborative; unconscious and conscious; bridging; text-connecting and gap-filling; coherence, elaborative, knowledge-based, and evaluative; and anaphoric, text-to-text, and background-to-text (Kispal, 2008).

Types of reading-related inferences include: referential, case structure role assignment, antecedent causal, superordinate goal, thematic, character emotion, causal consequence, instantiation noun category, instrument, subordinate goal action, state, reader’s emotion, and author’s intent (Graesser, Singer, & Trabasso, 1994).

Situation models, types of mental models, were devised to understand comprehension. These models are applicable to both story and social comprehension (Morrow, Bower, & Greenspan, 1989; Zwaan, Magliano, & Graesser, 1995; Zwaan & Radvansky, 1998; Wyer Jr, 2003).

Early research into machine story comprehension produced artificial intelligence systems which applied scripts, plans, plot units, and thematic structures. Examples of such systems include: SAM, PAM, FRUMP, and BORIS (László, 2008).

More recently, character networks can be extracted from stories (Labatut & Bost, 2019). In these dynamic networks, nodes correspond to characters and edges to the interactions between them. These nodes and edges can be mapped to embedding vectors (Lee & Jung, 2020; Hoang, Jeon, You, Yoon, Jung, & Lee, 2023).

Similarly, individuals in dynamic social-media networks can be mapped to embedding vectors (Pan & Ding, 2019; Hoang, Jeon, You, Yoon, Jung, & Lee, 2023).

Human lives can be viewed as sequences of events and represented in a way which shares a structural similarity with language. In the “life2vec” approach, resultant embedding spaces were found to be robust and highly structured (Savcisens, Eliassi-Rad, Hansen, Mortensen, Lilleholt, Rogers, Zettler, & Lehmann, 2023).

Characters in fictional and real-world stories could be mapped to corresponding “life2vec” vectors. These vectors would be updated as pertinent events occurred. With computational representations of situational contexts which include embedding vectors for characters, story-based contextual recommendations could be made for selected characters.

Systems capable of predicting stories’ trajectories and inferring characters’ mental states would make better story-based contextual recommendations (Gordon, Bejan, & Sagae, 2011; Chaturvedi, Peng, & Roth, 2017). The “life2vec” approach has shown promise with respect to both its predictive capabilities and its modeling of individuals’ personality nuances.

Inferring the goals and objectives of story characters and individuals will prove critical for contextually providing that content which is to be of the most use to them (Richards & Singer, 2001; Trabasso & Wiley, 2005). Computational approaches to these topics are explored in artificial intelligence with respect to robotic systems (Van-Horenbeke & Peer, 2021) and broader applications (Mao, Liu, Zhao, Ni, Lin, & He, 2023).

Beyond extracting character networks from stories, knowledge graphs could be extracted, mapped with embedding vectors, and subsequently utilized (Andrus, Nasiri, Cui, Cullen, & Fulda, 2022).

Narrative Psychology

Narrative psychology includes multiple parallel approaches: cognitive, psychometric, hermeneutic, scientific, and computational (László, 2008). Considered here are overlaps between artificial intelligence and those scientific and computational approaches of narrative psychology.

The contents and styles of the stories that individuals tell about their lives are of considerable importance. As story-based search and recommendation systems are constructed and continue to advance, opportunities for computer-aided and automated narrative coaching are expected to arise.

Narrative coaching works with coachees at three primary levels: (1) drawing on narrative psychology to understand and connect to the narrator, (2) drawing on narrative structure to understand and elicit the material in the narrated stories, and (3) drawing on narrative practices to understand and harvest the dynamics of the narrative field. The goal is to help coachees to forge new connections between their stories, their identity, and their behaviors in order to generate and embody new options in these three domains (Drake, 2010).

Mentoring and Coaching

Established theoretical models from mentoring and coaching will be of use for designing artificial intelligence systems which process stories or social media data to contextually search for and recommend items in a personalized manner for characters or individuals at scale.

Definitions of mentoring and coaching vary throughout the literature and have been the subjects of considerable debate (Passmore, Peterson, & Freire, 2016). For clarity, and for discussing artificial intelligence systems which can perform pertinent tasks, generic definitions of coaching and mentoring are offered here.

Mentoring is a relationship in which a mentor shares their knowledge, skills, and experience with a person, a mentee, to help them to progress.

Intelligent mentoring systems have been considered with respect to education, self-regulated learning, lifelong learning, career counseling, and beyond. With respect to mentoring beyond the scopes of educational courses or programs, challenges include the collection and integration of data from multiple sources to construct and maintain models of mentees (Kravčík, Schmid, & Igel, 2019).

Coaching is a form of human development in which a coach supports learners, clients, or coachees to achieve specific personal or professional goals by providing training and guidance. Coaching differs from mentoring by its focus on specific tasks or objectives, as opposed to a focus on more general goals or overall development. Applications of coaching include: business and executive, career, co-coaching, dating, education, financial, health and wellness, homework, life, relationship, religious, sports, vocal, and writing.

Individuals’ specific goals and objectives could be inferred by artificial intelligence systems and/or obtained through direct interactions using established theoretical models. Systems could interact with individuals using natural-language dialogues or by means of adaptive input forms. With detailed knowledge of individuals’ goals and objectives, intelligent coaching systems could better contextually recommend items for them.

The PRACTICE model details the following steps: problem identification, development of realistic goals, generation of alternative solutions, consideration of each solutions’ consequences, targeting of the most feasible solution, implementation of the chosen solution, and evaluation (Palmer, 2007). When goal-setting, SMART principles suggest that individuals’ goals should be specific, measurable, achievable, relevant, and time-bound (Doran, 1981).

To best obtain and maintain knowledge of individuals’ dynamic and unfolding goals and objectives, over time, frameworks for the design of intelligent coaching systems describe system attributes for developing strong and efficacious relationships: trust, empathy, transparency, predictability, reliability, ability, benevolence, and integrity (Terblanche, 2020).

Other models from positive psychological coaching can guide the design of intelligent coaching systems including: authentic happiness coaching, the flow-enhancing model, the co-active coaching model, positive organizational psychology, and the good work and good mentoring approach (Passmore, Peterson, & Freire, 2016).

Areas where intelligent coaching systems are expected to excel include evidence-based coaching and continual improvement. In these regards, multi-armed and contextual bandits address the primary difficulty of sequential decision-making under uncertainty, namely, the exploitation versus exploration dilemma. Exploitation involves choosing the best option based upon current knowledge of a system, while exploration involves trying out new options that may lead to better outcomes in the future at the expense of an exploitation opportunity. Applications of these techniques include: healthcare, e.g., clinical trials, recommender systems, information retrieval, and dialogue systems (Bouneffouf & Rish, 2019).


Mentors and coaches often offer advice to recipients. Items of wit and wisdom can be or can be contained in such messages of advice.

Research into advice can be organized into four paradigms: the message, discourse, psychological, and network paradigms. Each of these provides different insights about the characteristics, functions, and outcomes of advice. The message paradigm focuses on qualities of advice messages and on the effort to predict supportive outcomes for recipients, often between peers. The discourse paradigm provides insights into the structure and interpretation of advice in interactions. The psychological paradigm focuses on cognitive and emotional processes which predict the uses of advice in decision-making. The network paradigm highlights the utility of advice, often in organizational settings, as well as emergent global outcomes which arise from exchanges of advice (MacGeorge, Feng, & Guntzviller, 2016).

Wit and Wisdom

Beyond their uses in advice messages, items of wit and wisdom can amuse, inspire, motivate, and enlighten. They can express principles, values, and norms particular to cultures and eras (Hrisztova-Gotthardt & Varga, 2015).

These items should not only be indexed by their surface texts, but also by their interpretations. Items and their interpretations can be described with metadata schemas and interrelated to one another using formal ontologies.

Items can be mapped to embedding vectors (Rizkallah, Atiya, & Shaheen, 2021) and their interpretations can be mapped to vectors.

Interpretations can be diachronic, or dynamic, changing over the course of time. Interpretations can be isomorphic, identical in meaning to one another, or anti-isomorphic, opposite in meaning to one another (Yankah, 2015). Collections of items of wit and wisdom, then, can be paraconsistent.

Social Media

In the future, users of social media could be provided with means of browsing content pertinent to the situations described in their recent or selected posts, content aligned with their preferences and aesthetic tastes, while having the capability to provide feedback on the contextual recommendations and on the content recommended.

Artificial intelligence systems could provide multiple personas, each having different values, styles, or configurations with respect to content recommendation. In this way, individuals could browse and select from values, styles, and configurations using anthropomorphized personas. Opting into and out of content recommendation services could be as easy for individuals as friending and unfriending artificial intelligence personas.

At least initially, individuals might receive paginated lists of recommended items. Eventually, more advanced systems might be able to more intelligently sort items, refine items, and even decide upon single items.

Personalization and user modeling can be of use for enhancing contextual content recommendations. With personalization, systems can select and prioritize items aligned with individuals’ preferences and aesthetic tastes. Individual users, their preferences, and their aesthetic tastes can be represented using embedding vectors (Pan & Ding, 2019; Rizkallah, Atiya, & Shaheen, 2021).

Individuals should be able to provide feedback about contextual recommendations and the content recommended by means of using “like” buttons, upvoting mechanisms, input forms, or follow-up dialogues. Artificial intelligence systems could learn from and continuously improve using these and other sources of feedback.

While personalized content from artificial intelligence personas might be sent to individuals’ direct message inboxes, individuals should be able to easily repost or share these contents alongside any of their positive or negative comments, reactions, opinions, or evaluations.

Towards determining the value provided by contextually recommended content, artificial intelligence systems could observe individuals’ trajectories in embedding spaces after their encounters with recommended content. Encounters with recommended content could accompany individuals’ other social media data.

Research into moderating large language models is applicable to moderating story-based search and recommendation systems (Rebedea, Dinu, Sreedhar, Parisien, & Cohen, 2023). With respect to input moderation, for example, regions in situation spaces could be defined by system administrators as being inappropriate for their systems to provide content, advice or items of wit and wisdom, for.

Other Applications

In addition to their commercial applications, the technologies under discussion have applications to education, library and information science, knowledge management, and history.

With respect to education, contextually recommended items of wit and wisdom can provide educational value to individuals. Educational recommender systems have been previously explored for recommending academic advice, courses, educational programs, exams, learning resources, online learning opportunities, papers, pedagogical resources, professions, programming problems, study sequences or syllabuses, teaching practice resources, and schools or universities (Urdaneta-Ponte, Mendez-Zorrilla, & Oleagordia-Ruiz, 2021).

With respect to social-emotional learning and character education, representing learners’ paths as trajectories through embedding spaces could provide a new and powerful tool for understanding when best to use which pedagogical strategy.

With respect to library and information science, contextually recommended content, e.g., excerpts and quotations from literary works, could include hyperlinks to relevant books and materials.

With respect to knowledge management, organizations could index, search for, and retrieve content utilizing story-based contexts.

With respect to history, historians could contextually retrieve content, e.g., historical events and analogues, pertinent to contemporary societal-scale narratives.

Related Work

In a general sense, related work includes search engines, recommender systems (Aggarwal, 2016), reinforcement learning (Sutton & Barto, 2018), and vector database management systems (Pan, Wang, & Li, 2023).

Chatbots and dialogue systems are being researched for enhancing search and recommendation (Avula, Chadwick, Arguello, & Capra, 2018).

Improving recommender systems by incorporating social contextual information is being explored (Ma, Zhou, Lyu, & King, 2011) and so too are context-aware recommender systems for social networks (Suhaim & Berri, 2021).

Recommending quotations for dialogue systems and writing tasks are being researched (Ahn, Lee, Jeon, Ha, & Lee, 2016; MacLaughlin, Chen, Ayan, & Roth, 2021).

Research is underway into advice-related interactions between individuals and artificial intelligence systems (Liao, Oh, Feng, & Zhang, 2023).


Andrus, Berkeley R., Yeganeh Nasiri, Shilong Cui, Benjamin Cullen, and Nancy Fulda. "Enhanced story comprehension for large language models through dynamic document-based knowledge graphs." In Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 10, pp. 10436-10444. 2022.

Aggarwal, Charu C. Recommender systems: The textbook. Springer, 2016.

Ahn, Yeonchan, Hanbit Lee, Heesik Jeon, Seungdo Ha, and Sang-goo Lee. "Quote recommendation for dialogs and writings." In CBRecSys@RecSys, pp. 39-42. 2016.

Avula, Sandeep, Gordon Chadwick, Jaime Arguello, and Robert Capra. "Searchbots: User engagement with chatbots during collaborative search." In Proceedings of the 2018 conference on human information interaction and retrieval, pp. 52-61. 2018.

Bouneffouf, Djallel, and Irina Rish. "A survey on practical applications of multi-armed and contextual bandits." arXiv preprint arXiv:1904.10040 (2019).

Chaturvedi, Snigdha, Haoruo Peng, and Dan Roth. "Story comprehension for predicting what happens next." In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pp. 1603-1614. 2017.

Doran, George T. "There's a SMART way to write management’s goals and objectives." Management review 70, no. 11 (1981): 35-36.

Drake, David B. "Narrative coaching." In The complete handbook of coaching edited by Elaine Cox, Tatiana Bachkirova, and David Clutterbuck. p 120-131. SAGE. 2010.

Gordon, Andrew, Cosmin Bejan, and Kenji Sagae. "Commonsense causal reasoning using millions of personal stories." In Proceedings of the AAAI Conference on Artificial Intelligence, vol. 25, no. 1, pp. 1180-1185. 2011.

Graesser, Arthur C., Murray Singer, and Tom Trabasso. "Constructing inferences during narrative text comprehension." Psychological review 101, no. 3 (1994): 371.

Hoang, Van Thuy, Hyeon-Ju Jeon, Eun-Soon You, Yoewon Yoon, Sungyeop Jung, and O-Joun Lee. "Graph representation learning and its applications: A survey." Sensors 23, no. 8 (2023): 4168.

Honeck, Richard P. A proverb in mind: The cognitive science of proverbial wit and wisdom. Psychology Press, 2013.

Hrisztova-Gotthardt, Hrisztalina, and Melita Aleksa Varga, eds. Introduction to paremiology: A comprehensive guide to proverb studies. De Gruyter Open, 2015.

Kispal, Anne. Effective teaching of inference skills for reading: Literature review. National Foundation for Educational Research. The Mere, Upton Park, Slough, Berkshire, SL1 2DQ, UK. 2008.

Kravčík, Milos, Katharina Schmid, and Christoph Igel. "Towards requirements for intelligent mentoring systems." In Proceedings of the 23rd International Workshop on Personalization and Recommendation on the Web and Beyond, pp. 19-21. 2019.

Labatut, Vincent, and Xavier Bost. "Extraction and analysis of fictional character networks: A survey." ACM Computing Surveys (CSUR) 52, no. 5 (2019): 1-40.

László, János. The science of stories: An introduction to narrative psychology. Routledge, 2008.

Lee, O-Joun, and Jason J. Jung. "Story embedding: Learning distributed representations of stories based on character networks." Artificial Intelligence 281 (2020): 103235.

Liao, Wang, Yoo Jung Oh, Bo Feng, and Jingwen Zhang. "Understanding the influence discrepancy between human and artificial agent in advice interactions: The role of stereotypical perception of agency." Communication Research (2023): 00936502221138427.

Ma, Hao, Tom Chao Zhou, Michael R. Lyu, and Irwin King. "Improving recommender systems by incorporating social contextual information." ACM Transactions on Information Systems (TOIS) 29, no. 2 (2011): 1-23.

MacGeorge, Erina L., Bo Feng, and Lisa M. Guntzviller. "Advice: Expanding the communication paradigm." Communication yearbook 40 (2016): 239-270.

MacLaughlin, Ansel, Tao Chen, Burcu Karagol Ayan, and Dan Roth. "Context-based quotation recommendation." In Proceedings of the International AAAI Conference on Web and Social Media, vol. 15, pp. 397-408. 2021.

Mao, Yuanyuan, Shuang Liu, Pengshuai Zhao, Qin Ni, Xin Lin, and Liang He. "A review on machine theory of mind." arXiv preprint arXiv:2303.11594 (2023).

Mieder, Wolfgang, ed. Wise words: Essays on the proverb. Routledge, 2015.

Morrow, Daniel G., Gordon H. Bower, and Steven L. Greenspan. "Updating situation models during narrative comprehension." Journal of memory and language 28, no. 3 (1989): 292-312.

Palmer, Stephen. "PRACTICE: A model suitable for coaching, counselling, psychotherapy and stress management." The Coaching Psychologist 3, no. 2 (2007): 71-77.

Pan, James J., Jianguo Wang, and Guoliang Li. "Survey of vector database management systems." arXiv preprint arXiv:2310.14021 (2023).

Pan, Shimei, and Tao Ding. "Social media-based user embedding: A literature review." arXiv preprint arXiv:1907.00725 (2019).

Passmore, Jonathan, David Peterson, and Teresa Freire, eds. The wiley blackwell handbook of the psychology of coaching and mentoring. Nashville, TN: John Wiley & Sons. 2016.

Rebedea, Traian, Razvan Dinu, Makesh Sreedhar, Christopher Parisien, and Jonathan Cohen. "Nemo guardrails: A toolkit for controllable and safe llm applications with programmable rails." arXiv preprint arXiv:2310.10501 (2023).

Richards, Eric, and Murray Singer. "Representation of complex goal structures in narrative comprehension." Discourse Processes 31, no. 2 (2001): 111-135.

Rizkallah, Sandra, Amir F. Atiya, and Samir Shaheen. "New vector-space embeddings for recommender systems." Applied Sciences 11, no. 14 (2021): 6477.

Savcisens, Germans, Tina Eliassi-Rad, Lars K. Hansen, Laust H. Mortensen, Lau Lilleholt, Anna Rogers, Ingo Zettler, and Sune Lehmann. "Using sequences of life-events to predict human lives." Nature Computational Science (2023): 1-14.

Suhaim, Areej Bin, and Jawad Berri. "Context-aware recommender systems for social networks: review, challenges and opportunities." IEEE Access 9 (2021): 57440-57463.

Sutton, Richard S., and Andrew G. Barto. Reinforcement learning: An introduction. MIT press, 2018.

Terblanche, Nicky. "A design framework to create artificial intelligence coaches." International Journal of Evidence Based Coaching & Mentoring 18, no. 2 (2020).

Trabasso, Tom, and Jennifer Wiley. "Goal plans of action and inferences during comprehension of narratives." Discourse processes 39, no. 2-3 (2005): 129-164.

Urdaneta-Ponte, María Cora, Amaia Mendez-Zorrilla, and Ibon Oleagordia-Ruiz. "Recommendation systems for education: Systematic review." Electronics 10, no. 14 (2021): 1611.

Van-Horenbeke, Franz A., and Angelika Peer. "Activity, plan, and goal recognition: A review." Frontiers in Robotics and AI 8 (2021): 643010.

Wyer Jr, Robert S. Social comprehension and judgment: The role of situation models, narratives, and implicit theories. Psychology Press, 2003.

Yankah, Kwesi. "Do proverbs contradict?." In Wise words: Essays on the proverb, pp. 127-142. Routledge, 2015.

Zwaan, Rolf A., Joseph P. Magliano, and Arthur C. Graesser. "Dimensions of situation model construction in narrative comprehension." Journal of experimental psychology: Learning, memory, and cognition 21, no. 2 (1995): 386.

Zwaan, Rolf A., and Gabriel A. Radvansky. "Situation models in language comprehension and memory." Psychological bulletin 123, no. 2 (1998): 162.