Search, sense making and learning: closing gaps - Leitura de Artigo

Marchionini, G. (2019), "Search, sense making and learning: closing gaps", Information and Learning Sciences, Vol. 120 No. 1/2, pp. 74-86. https://doi.org/10.1108/ILS-06-2018-0049 

The primary claim of this paper is that electronic information tools and environments have increased the number and intensity of overlaps between search and learning, and in turn made search and learning less discrete and more seamless. Additionally, it claims that the sense making process – understanding, interpreting and reflecting on information units

[Aprender enquanto busca, usar o que se sabe para buscar. Mas a busca dispensa o uso da memória?] 

Next, sense making is discussed as a component of both search and lerning. Third, learning is discussed as a set of physical, mental, and affective changes that entail retention, and this discussion is considered with respect to search and sense making.

[Aprender implica em reter, Look Up é pontual pq a informação recuperada pode ser descartada depois]

I argue that simple lookup type search has become an ‘automatic’ building block for more complex and persistent everyday goals, including learning. Although retrieval research and development progress has been remarkable, much remains to be done as we expect to find and use information for the full range of complex human activities.

[Busca com um propósito mais amplo]

In this paper, search is considered in this broad sense with information retrieval and information seeking as particular types of search. The information retrieval (IR) research community focuses its interest on systems and services that support the search for information, and historically on the search for information that exists – thus can be retrieved.

[Lacunas de conhecimento, não é possível recuperar o que não está registrado e também o que ainda não se sabe]

Information seeking is a broader term than retrieval because it does not assume that the needed information exists and focuses on mental and physical activities that may take place over time or in collaboration.

[Quem busca no KG pode colaborar com quem cria o KG. A busca pode aponta lacunas no KG que é incompleto.]

The boundary conditions of retrieval (i.e. extant, discrete units) have served humanity well because they allowed researchers and developers to attend to the challenges of representation of existing information, what humans articulate as their current information need (e.g. via query), and the challenges of matching these representations.

[Como representar o conhecimento, como representar a necessidade de informação e como fazer o match entre ambos]

Since the 1960s, there have been three dramatic improvements in IR research and development.

[Evolução IR em 3 fases: Indexar, Ranquear, Contextualizar]

1) These systems dramatically improved human ability to retrieve existing information items across multiple sources and with more discrete control over index assignments.

2) Larger storage allowed information item indexing to move beyond small controlled vocabularies to abstract or full text indexing, and improved computation allowed systems to rank order retrieved items according to similarity based on word occurrence statistics or probabilities. 

3) Search engines take into account searcher characteristics (e.g. location, search history) as well as collective context (e.g. topics that are trending). Search engines apply sophisticated natural language techniques that take advantage of enormous corpora of text from different contexts and in different languages and equally importantly, use logs of billions of queries to tune a searcher’s query statement to a probabilistic context that returns excellent results for most searches[1].

[Contexto que atende a maior parte da consultas mas não é possível contextualizar a consulta que será submetida para atender a outros cenários, cauda longa]

Today, searchers type or speak words, phrases or questions are aided by autocomplete (anticipatory) assistance, and immediately are provided with sets of potentially pertinent items ordered in presumably useful ways.

[Antecipar necessidades de informação monitorando aplicativos]

Information seeking

I have argued that information seeking aided by search systems is a broader and more human-centered concept than retrieval (Marchionini, 1995). The information seeking process includes a set of subprocesses with varied transition probabilities among the subprocesses (recognize and accept the information need, define the problem/need, select source, formulate query, execute query, examine results, extract information and reflect/stop).

[Humano e não máquina, Look Up pode ser usado pela máquina]

Much more time and effort are given to examination of results with today’s search systems because knowledge creation and accessibility continues to both accelerate and diversify.

[Conhecimento humano gerado pela análise dos resultados]

Information seeking research has tended to focus on the human factors and the contextual factors of search. Researchers have conducted investigations of the human needs that motivate search.

[Contexto é considerado na busca]

The most significant development in information seeking research over the last two decades has been the trend toward interactive search.

[Não é atômico]

White and Roth (2009), for example, have demonstrated the range and efficacy of exploratory search. Interactive search incorporates different system, human and contextual factors that change the search process and experience.

[Como demonstrar eficácia? ]

[White, R. W., Roth, R. A. (2009). Exploratory Search: Beyond the Query-Response Paradigm. [San Rafael, Calif.]: Morgan & Claypool Publishers. ISBN: 978-1-59829-783-6 ]

The human factors include shifting search from a discrete turn-taking task that mainly entails cycles of high-cognitive load planning and examination to rapid cycles of action that entails low-cognitive load recognition subtasks and the examination of search results via a wide variety of display

[Interativo e sem muito planejamento, ciclos mais rápidos]

Erdelez (2005) and her colleagues investigate the roles of serendipity and information rich environments play under the rubric of information encountering.

[Serendipidade: descobertas inusitadas]

We are increasingly engaged in what might be called anticipatory search – default settings on our information devices that provide results that we call alerts or recommendations. ... anticipates information needs and invites context-specific results, ... our anticipated searches become even more pervasive so that we need new strategies for filtering and managing the streams of context-relevant information spawned by our cascading default settings (anticipatory queries).

[Recomendação: foco na "melhor" experiência do usuário. Causaria sobrecarga?]

The figure situates search, learning, information seeking, and information retrieval as overlapping activities with more overlap between information seeking and learning than between learning and information retrieval, with each having distinct characteristics not found in the others. 

[IR está parcialmente contido em IS e ambos estão totalmente contidos em Search. Search está parcialmente contido em Learning. Existem outras formas de aprender e IR atenderia a outras motivações de busca que não sejam exploratórias.]

I have also reserved information seeking for humans and allow retrieval to be a human or a machine activity.

[Diferença crucial]

Sense making

... cognitive processes that humans apply to identify representations for complex information tasks, encode data into these representations and iteratively modify those representations to minimize cognitive effort and maximize task solution effectiveness (Russell et al., 1993). This third meaning mainly considers search as a subtask of sense making, which in turn is a subtask within larger problem solving or decision making tasks.

[A forma como a informação é representada interfere no esforço cognitivo aplicado. Sensemaking é o processo pelo o qual o ser humano cria sentido acerca do mundo que o rodeia para tomar decisões. E a busca ajuda ao ser humano a encontrar informações que descrevam esse mundo.]

workshop at the annual ACM CHI Conference that included 25 peer reviewed papers

[ACM Conference on Human Factors in Computing Systems (CHI)]

Here, I take sense making to be an interpretive mental activity that connects existing knowledge (what we have learned) to the immediate stimuli that occur during search (e.g. result page characteristics, query suggestions and other systemfeedback).

[Outra definição para sense making] 

Sense making typically does lead to learning, which, as discussed below, demands retention.
When humans encounter or are presented with information we use a variety of mental skills to act or move on. Some of these skills include tolerance for ambiguity or tolerance to ignore information; maintaining focus; visually scanning key features (e.g. words, facial expressions, motion); reading or studying details (decoding), comparing features to personal knowledge (identifying, inferring) and leveraging external resources to clarify (e.g. reference materials such as dictionaries, Wikipedia, asking other people, related search results).

[Hiper KG permitiria recuperar afirmações contraditórias e com diferentes níveis de acurácia.]

Learning

Educational theorists continue to develop cognitive models of learning but often aim to connect them to neuroscience research (e.g. enhanced cognitive function), pharmacology (memory or skill enhancement) or computational aids (e.g. intelligent tutors, immersive environments, intelligent games). Here, we focus on the relationships between learning and search. 

[Aprender estaria intimamente ligado a Busca em função da disponibilidade de material digital]

Vakkari (2016) in his review of search as a learning process discusses the importance of including the processing of search results (either to continue the next iteration of search or to begin to use the results to meet the information need) as one of the characteristics that brings search closer to learning. He also points out the need for process-oriented metrics to assess search rather than relying on traditional outcome measures like recall and precision.

[Existiriam essa métricas para avaliar aprendizado?]

I suggest that when sense making tools are added to retrieval tools in today’s highly interactive environments for exploration and learning, a new kind of thinking emerges.

[Nesse pensamento seia possível questionar o que está sendo apresentado como resultado]

What do we learn?

Humans are learning systems. We acquire facts. We develop skills and strategies. We construct knowledge networks. We develop preferences and attitudes. We reflect on these processes and outcomes. Clearly, each of these activities can take place at micro, intermediate or macro levels of complexity.

[Primeira dúvida é se sempre adquirimos só fatos, podemos aceitar as afirmações contextualizadas sem colocar nelas um valor de Verdade Absoluta? Outra é a rede de conhecimento, tudo está interligado]

It is obvious how search is critical to fact acquisition. Fact acquisition is most often a discrete act and although we certainly acquire facts incidentally as we live our lives, efforts to acquire particular or immediately pertinent facts are strongly aided by search strategies and systems. Sense making may aid retention by forming linkages to other facts and to existing mental models. Performance for fact acquisition is assessed by recall on demand.

[FATOS]

Skills and strategies are developed rather than acquired. They require practice over time. ... Retention is determined by practice and performance by execution in real or simulated conditions. 

[Ensinar ao usuário a avaliar o contexto além das afirmações. A presença do contexto nas repostas estimularia o usuário a desejar que todas as respostas fossem completamente contextualizadas ou seria cansativo?]

Knowledge construction implies creating arrays of relationships between objects, agents and actions that are situated in a context. ... Specific retrieval acts are necessary but not sufficient for knowledge construction. Searching over time and possibly in collaboration with others aids knowledge construction, maintenance and application. Sense making is essential to evaluate intermediate results, reconfigure representations and identify relationships among entities in the network. Because knowledge networks are highly fluid and evolve over time, it is more useful to consider richness and depth rather than retention. I have used the idea of noumenal clouds that form and change to describe the concepts within a knowledge network (Marchionini, 1995). Learning performance is associated with application and transfer rather than recall or execution.

[Grafos de Conhecimento para representar]

It is not clear how search can help, although the broader and repeated search activity might be part of one’s quest for experiences that influence attitudes. 

[A atitude de não aceitar Verdades Absolutas mas entender que podem existir diferentes contextos para uma afirmação ser verdadeira ou falsa. E que em determinadas situações é necessário agir / tomar decisão mesmo com as informação não estando completamente contextualizadas e isso traz incerteza ao processo.]

Learning can be intentional or incidental and most of our concerns in the educational and information literatures are with intentional learning. Incidental learning is becoming increasingly important in our highly mediated world and overlaps with serendipity during search. Intentional learning can be directed (supervised) or self-directed (unsupervised). Directed learning is most obviously exemplified in formal schools, whereas self-directed learning is most obviously exemplified in activities people do in libraries or “on the job”.

[Aprendizado supervisionado e não supervisionado tem o mesmo sentido de ML]

Closing the gap

Figure 2 depicts the relationships among search, sense making, and learning. In this paper, I emphasize search as a subtask for learning and the arrowhead pointing to learning is much larger than the arrowhead pointing back to search. This is meant to show that people do learn to search while engaged in the search process, however, in general search to support learning is a much more common human activity. Sense making is depicted as an intermediary between search and learning where a mental model for the object of learning is identified and then results mapped and contextualized to that mental model. Also, what we “know” (have learned) may assist search progress by “making sense” of the stimuli spawned by the search process.

[O modelo de confiança faria parte desse modelo mental?]

Easy access to much of the world’s knowledge provides unprecedented opportunity for human problem solving, decision making, and learning – potentially augmenting our intellects. However, access is not sufficient for intelligence – we must manage different search strategies (e.g. query, browse, select, set defaults) and control mechanisms (e.g. type, click, speak, gesture) and filter, organize, evaluate and interpret streams of results.

[Além do acesso é preciso ter capacidade de criticar o que é apresentado]

As we incorporate immediate and effortless search into our daily lives as an automatic process, our behaviors based on immediate recall of facts evolve; however, we must consider alternative pathways for search so that our needs for complex or highly specialized knowledge and the corresponding complexity of search strategies do not atrophy. 

[Reduzir o esforço de encontar (eficiência de IR) mas não de analisar]

For example, the effects of overloads caused by anticipated search streams and managing the inevitable collections of consequent filters in real time may change how we search, make sense and learn as we increasingly augment ourselves with digital tools and services. Additionally, we must encourage and teach new kinds of critical evaluation strategies to cope with adversarial, biased, or manipulative
content and search mechanisms. 

[Por isso trabalhar com afirmações contextualizadas, explicitar o que pode estar enviesado]

I suggest that we need to build bridges to the neuroscience and behavioral science research communities so that we can understand the biological, affective and behavioral consequences of ubiquitous search on the whole individual and society at large.

[Espaço para cooperação com outras disciplinas]

Just as spreadsheets enabled “what if” thinking to become easily and pervasively available, today’s search environments allow large portions of humanity to engage in exploratory forays into concepts, events and people in unprecedented ways.

[Simulações com planilhas]

It is important that people understand that search results are “accurate” at probability levels rather than absolute levels. The many adjustments search engines make to optimize search results entail many computational estimates that typically do optimize search results, however, not always. Because these underlying adjustments are either proprietary or include many “hidden layers” of computation, we are not able to get evidence of why search results are given.

[A ordenação de máquinas de busca não é clara para o usuário e depende de muitas variáveis.A ordem não é JUSTIFICADA mas influencia no aprendizado pq as pessoas tendem a olhar os primeiros resultados. Se os critérios fossem explícitos e controláveis pelo usuário a busca poderia ser mais eficaz em resolver a necessidade de informação]