CollectiveTeach: Crowdsourcing Lesson PlansAshwin VenkataramanRishabh RanawatNew York [email protected]. of Curriculum and InstructionCollege of Education, UT [email protected] ChenSrikanth JagabathulaLakshminarayananSubramanianDept. of Computer ScienceNYU Abu [email protected] Dept.NYU Stern Business [email protected] of quality textbooks and good educational resources is a wellknown problem in developing regions. In this paper, we describethe design of CollectiveTeach, a web platform that aims to integrate rich educational content into an inquiry-based framework,viz. the 5E learning model, for generating web-annotated lessonplans for school and college teachers in developing countries. Giventhe wealth of educational resources on the Web, CollectiveTeachhelps teachers to author new lesson plans through a simple webinterface allowing them to easily search, select, order and collecteducational content relevant to the topics they wish to teach. Thispaper describes our experiences building two versions of the CollectiveTeach platform. The initial platform was tested via a userstudy with a cohort of 19 teachers in Ghana and the learnings ofthis user study were incorporated in a second, improved version ofCollectiveTeach; the current prototype of CollectiveTeach was evaluated using human experts for computer science subjects coveredin standard undergraduate curriculum.ACM Reference format:Ashwin Venkataraman, Rishabh Ranawat, Sepehr Vakil, Jay Chen, SrikanthJagabathula, and Lakshminarayanan Subramanian. 2016. CollectiveTeach:Crowdsourcing Lesson Plans. In Proceedings of ACM Conference, Washington,DC, USA, July 2017 (Conference’17), 9 pages.DOI: 10.1145/nnnnnnn.nnnnnnn1Sepehr VakilDept. of Computer ScienceNYU [email protected] the last several decades, there has been great progress inintegrating inquiry-based teaching strategies into the classroom [15,16]. Simultaneously, there has been an explosion of online webbased learning materials and initiatives to introduce computers intoclassrooms [9, 25]. Unfortunately, teachers have only a few toolsfor making productive use of online content within inquiry-basedframeworks for teaching and learning. This problem is worse indeveloping regions where lower levels of technological literacyPermission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for third-party components of this work must be honored.For all other uses, contact the owner/author(s).Conference’17, Washington, DC, USA 2016 Copyright held by the owner/author(s). 978-x-xxxx-xxxx-x/YY/MM. . . 15.00DOI: 10.1145/nnnnnnn.nnnnnnnDept. of Computer ScienceNYU [email protected] severe challenges to teachers seeking to incorporate webresources in their teaching materials and classroom instruction.This paper presents the design, implementation and early deployment experiences of CollectiveTeach, an online lesson plangeneration platform that enables teachers to collectively organizeweb-based educational resources within an inquiry-based framework for teaching and learning. The design of CollectiveTeachdraws inspiration from the 5E model [10] based on the educationalphilosophy and psychology of Johann Herbart [23], which has along history in educational theory grounded in ideas of Piaget andDewey [29]. The 5E model for preparing a lesson plan comprisesof 5 stages: Engage, Explore, Explain, Elaborate and Evaluate. CollectiveTeach enables teachers to express the key topics in theirlesson plan across these 5 stages in the 5E model and translates theteacher specifications to a list of appropriate and targeted (web-)search queries. CollectiveTeach presents a simple interface for teachers to inspect the top-ranked search results in each stage to selectand order relevant web content for each part of their lesson plan;collectively, the user-chosen content coupled with the corresponding URLs forms a web-annotated lesson plan created by a teacher fora specific educational class. In addition, CollectiveTeach supportssearch and upload capabilities that enable teachers to incorporatemultimedia rich content including video, images, presentations,documents and domain-specific educational resources without being overly complex in terms of user interface elements or additionalfeatures.We present our experiences developing two versions of the CollectiveTeach platform across a multi-year effort. In 2013, we developed an early prototype version of the CollectiveTeach platformand evaluated the effectiveness of the platform based on interactions with 19 K-12 mathematics and science teachers in Ghanaduring two week-long workshops. We observed that the 5E modelnaturally fit the existing instruction style of most teachers whoparticipated in our user study. Results from the study show thatteachers broadly found CollectiveTeach to be a highly effective toolfor both creating inquiry-based lesson plans and also integratingweb-based multimedia content in mathematics and science classes.Despite the fact that the teachers rarely used web content whencreating existing lesson plans and most of them were only formallyintroduced to the 5E model in our training sessions, we found thatthe teachers were able to quickly and easily use our platform.

Conference’17, July 2017, Washington, DC, USAThe second version of the CollectiveTeach platform addressedmany of the usability and content organization challenges that werepresent in the first version. The current prototype version supportsa much richer set of functionalities and enables easy creation oflesson plans for new subjects within a few minutes. We evaluatethe effectiveness of the second prototype version of CollectiveTeachusing a human expert-based rating approach for popular computerscience subjects to determine the utility and coverage of contentshown by the generated lesson plans. In summary, we believe thatCollectiveTeach can be adopted as an effective platform for enablingeasy creation of web-based lesson plans using online educationalresources.2RELATED WORKA growing community of scholars including educational researchersand computer scientists are interested in how technology can support new modes of learning and instruction. Research on the designof learning environments that leverage the rapid advancement ofinformation technologies can be found in programs such as theWeb-based Inquiry Science Environment (WISE) [4] and LearningTechnologies in Urban Schools (LeTUS) among others [2]. Manyof these projects explore the affordances of specific technologiessuch as visualizations, models, and data probes for inquiry-basedlearning. In addition to student supports for learning, the WISEprogram mentioned above also provides teacher supports for authorship and customization of curriculum [31, 32]. A product ofover 20 years of research on learning with educational technologies,WISE is an example of a powerful educational tool based on traditions within constructivist philosophy, in particular the KnowledgeIntegration framework. The design of the CollectiveTeach platformwas strongly informed by projects such as WISE, in particular theapproach of incorporating constructivist perspectives on learningwithin the design of the platform itself. However, a limitation ofprograms such as WISE within the developing region context iscomplexity. The WISE tool includes advanced features that maypose significant challenges for teachers (and students) who havelimited experience with computers and learning technologies [28].In conjunction with the growth of the Web, there has been asimilar growth in number of projects that have leveraged technology and web-based resources for education [2, 3, 19], especially fordeveloping regions [1, 6, 11, 17, 19, 22, 30, 34]. The Hole-in-the-Wallproject showed that exposure of web-based educational resourcesto under-privileged students without any formal guidance on innovative learning platforms can indeed produce positive learningoutcomes [30]. Digital Study Hall is another successful project thathas used participatory videos to enhance rural school education[34]. There have also been studies which have analyzed the meritsof Powerpoint slides [27] and video materials [3, 21] as teachingaids. Recent work looking at textbook content in developing regionsexpose the limitations of existing educational resources [14, 20] andpropose methods to automatically augment existing course materials with online content [6]. The basic rationale behind these worksis similar to the focus of our paper in that the web has a wealth ofeducational information that can be used for enhancing classroomcontent. Simple tools and interfaces can dramatically lower thebarriers to entry [7]. Chakraborty et al. found that there was sufficient quality content on the web and that many web resources wereA. Venkataraman et al.ideal for activity-based learning [11]. The major drawbacks of suchsystems is that despite the large amount of educational content,finding the most relevant content for a particular class may take along time. Therefore, filtering the appropriate content and helpingto organize the content for instructors is a central problem that isaddressed in the design of CollectiveTeach .3THE COLLECTIVETEACH PHILOSOPHYThe basic building block of CollectiveTeach is the concept of anatomic learning unit (ALU), which is a concise and compact educational material (spanning less than a few pages of textbook content)that primarily discuss a single concept. This is similar to the conceptof “key section” introduced by Agrawal [5]. Given our focus onweb based educational resources in the CollectiveTeach framework,an atomic learning unit can refer to the content in a single webpage, relevant pages within an online document, or user postedcontent within the CollectiveTeach system. An atomic learning unitcan refer to different types of content that pertain to the conceptsuch as: (a) textual content that explains the relevant concept; (b) acollection of problems and solutions; (c) multimedia content (video,images); (d) user-uploaded content (pdfs, powerpoints, etc.). Inthe CollectiveTeach framework, an ALU refers to a single educational resource link on the Web or a user-uploaded file that providesconcise educational material relating to a given concept.Consider a lesson to represent the material covered in a singlelecture in a traditional educational setting. We consider a lesson tobe a collection of related concepts. In the CollectiveTeach framework, a lesson plan is an ordered collection of atomic learning unitsthat is necessary to teach the concepts in a lesson. ALUs serve as alogical partition of the instructional content of a lesson with thepurpose of effectively communicating concepts to students.The CollectiveTeach lesson plan creation framework is directlyinspired by the 5E model in the education pedagogy literature [24].Building off prior learning cycle models, the Biological SciencesCurriculum Study (BSCS) 5E model is based on educational researchtracing back to the early constructivist perspectives of Piaget andDewey. As a robust philosophical tradition within education, constructivism is fundamentally concerned with how knowledge isconstructed and therefore has significant implications for theories of instruction and curriculum development. The 5E model isone of the best known within this tradition. The BSCS 5E modelcontributes two additional stages, engagement and evaluation, tothe Science Curriculum Improvement Study (SCIS) learning cyclewhich was comprised of three stages: exploration, invention, anddiscovery [10]. Not intended as a linear formula but rather a guideto structure activity in a way that centers students in the learning process, the 5E learning cycle consists of five stages: Engage,Explore, Explain, Elaborate, and Evaluate.Engage: The first stage in the learning requires engaging thestudents. Engagement generally involves connecting the topic ofinstruction to the lived experiences of students through culturallyrelevant pedagogical practices. This may include showing videos,open-ended discussions, freewrites, or class debates designed to“hook” students into the topic of instruction [10].Explore: Next, through hands-on activities, labs, or class discussions, students are guided to explore a topic that may have emerged

Conference’17, July 2017, Washington, DC, USAin the engagement phase, or may relate more centrally to the topicof instruction. Explore activities generally help the teachers gain asense of student prior knowledge, which skillful teachers will takeinto account as they unfold the rest of their lesson.Explain: The new ideas generated, and the questions raised during exploration activities will help teachers target their instructionduring the explain stage. Often during this step in the learningcycle, teachers will confirm student ideas or help clarify studentmisconceptions revealed during earlier stages in the process. Concrete activities usually involve direct instruction and presentationof scientific terms or ideas.Elaborate: The elaboration stage is intended for deeper inquiryinto the topic of instruction by challenging students with complexproblems and demonstrating real-world applications of the conceptsdiscussed. Common activities may include group problem-solvingchallenges and group discussions connecting topic of instructionto real-world applications.Evaluate: Finally, the evaluation stage provides an opportunityfor teacher and students to assess understanding and conceptualmastery of the information provided throughout the course of theclass.the option to “add web resources” to enrich or support the specified activities, by choosing the type of resource (web, video, image,etc.) as well as a subset of common keywords that can enhance thesearch query and increase the likelihood of returning web resultsthat are both relevant and of high quality. In our preliminary testswe found that for specific stages in the 5E model, adding specialcommon keywords like “problems”, “questions”, “examples”, “applications” etc. can significantly enhance the quality of the searchresults. After teachers have completed outlining their lesson planacross the 5E stages, CollectiveTeach 1.0 converts the user inputinto a set of appropriate search queries and returns the top 3 searchresults for each “resource” requested, in a common web interfacefront-end. Finally, our platform provides a simple interface to enable teachers to sift through the collection of all search results andselect which results have the relevant content they are looking for.For multimedia- rich content, CollectiveTeach 1.0 provides a directsnapshot of the content to enable faster selection from the user.Since the result pages are organized across the different stages, thechosen collection of search results form the web-annotated lessonplan constructed by the teacher.44.1COLLECTIVETEACH 1.0CollectiveTeach 1.0 was designed as a simple prototype systemthat integrated easily into the existing educational ecosystem withminimal training, cost or maintenance. Our goal was to introducea platform that would speak to the needs of our target population(teachers) using a collaborative design approach [8]. The CollectiveTeach 1.0 platform enables a teacher who intends to teach aclass on a given topic to easily find relevant online educationalresources and create a Web-annotated lesson plan for her class. Thedesign of CollectiveTeach 1.0 assumes that the teacher has a roughflow of the list of topics she intends to cover in her class and thelesson plan, in essence, is a set of curated web contents carefullychosen (by the teacher) to best fit the material covered in class.For example, a Biology teacher teaching human anatomy couldsignificantly benefit from a wealth of images/videos on the webdescribing anatomy of the human body as an educational resourcefor both teaching her class more effectively as well as an additionallearning guide for her students. There are two significant researchchallenges that need to be addressed to design a system that meetsthe above objectives: (1) Given the large volume of educational information on the web, how can teachers identify relevant and highquality materials that will support student learning in the classroom?Embedded within this challenge is the task of formulating the righttype of search query that will return relevant content among thetop few search results. (2) Given a rough outline of how a teachermay want to organize her materials for a class, how can she assimilateall the relevant web information she finds into a coherent lesson plan?To address both these challenges, CollectiveTeach 1.0 leveragesthe 5E instructional model and uses the following multi-step approach to enable teachers to create their lesson plans. First, CollectiveTeach 1.0 provides a simple HTML form interface that guidesteachers to organize their lesson plan outline within the frameworkof the 5E model. For each stage in the 5E model, it prompts theteacher to describe the instructional activities they intend for students to interact with in that stage. Additionally, teachers are givenEarly User StudyWe tested the effectiveness of CollectiveTeach 1.0 with mathematicsand science teachers in Accra, Ghana. We recruited 19 teachers froma school district in the Accra region comprising primary, middleschool and high schools, and both private as well as governmentschools. The participants had between 6 and 29 years (mean of 14.8)of teaching experience. Five participants were female (26%). Fifteenparticipants had a computer at home (79%), and 10 had Internet athome (52%). Teachers were invited to participate in two workshopsdesigned to explore the role of technology in education with a focuson math and science subjects.In the first workshop, participants completed a 20-minute presurvey designed to better understand their current computer andtechnology skills along with their approach to creating lesson-plans.In addition, a 30-minute orientation to the 5E model and tutorialof the CollectiveTeach 1.0 platform was conducted. Then, the participants used the CollectiveTeach 1.0 platform for 45 minutes tocreate web-based lesson plans and finally, completed a post-surveyfocused on gaining feedback on specific features of the platform.Approximately a week after the first workshop, 10 participantswere invited again based on the quality of lesson plans that theycreated to participate in the second workshop which consisted ofthe following 3 steps:(1) Participants created two new lesson plans using the platform based on their existing curriculum materials(2) Participants assessed their own lesson plans on a Likertscale rating system across 4 dimensions:(a) Consistency of lesson plan with 5E instructional framework(b) Skillful integration of web-content into lesson plan(c) Search results match lesson plan(d) Overall rating of lesson plan(3) Participants rated the other participants’ lesson plans acrossthe same dimensions

Conference’17, July 2017, Washington, DC, USAA. Venkataraman et al.Table 1: Post-survey results (Workshop #1). (Likert scale1 Strongly Disagree to 5 Strongly Agree)I usually use web resources for lesson planningThe 5E model was helpfulI thought the platform was easy to useThe platform returned useful search resultsWeb resources were usefulOverall, the platform was helpful for the tasksI would use this platform again2. process of participants assessing their own lesson plans, aswell as lesson plans of other participants informs our analysis ofthe success of the CollectiveTeach 1.0 platform presented below.Survey Results. After participants completed workshop #1 ofthe user study, they were given a post-survey. Table 1 summarizesthese results. These figures indicate that overall participants foundthat the 5E model was helpful, the platform was easy to use, webresources returned were useful, and the platform was helpful forthe lesson plan construction process.Usefulness of the Platform. During the second workshop,after participants constructed two lesson plans based on their owncurriculum materials, they were asked to assess their lesson plansacross several dimensions relating to overall quality. Each of theten participants who participated in the second workshop createdtwo new plans, resulting in 20 self-rated lesson plans. Of these 20self-assessments, 7 were marked “strongly agree” and 13 “agree”when presented the following statement:“I believe this lesson plan is created according to the principles of5E, resulting in engaging activities that emphasize student-centeredlearning and conceptual mastery of the material.”When participants rated each other’s lesson plans using the sameprompt, the results were nearly identical, with 6 marks for “stronglyagree”, 12 for “agree”, 1 for “disagree” and one left unscored. Overall,these results indicate that participants expressed confidence that theCollectiveTeach 1.0 platform assisted the process of mapping theirexisting curricular materials into a cohesive lesson plan, withinthe inquiry-based 5E framework. This matches closely with theresults from Workshop #1 that indicated our participants found theplatform useful for completion of lesson plan creation tasks.4.2 Adoption of this tool not feasible without an increase inteacher knowledge. (structural) To be able to adopt this model here will require massiverefurbishment of our computer labs (structural)5COLLECTIVETEACH 2.0Based on the limitations identified above, we enhanced CollectiveTeach 2.0 with the following five features—(1) Query formulation, (2) Reordering, editing and searching,(3) Automated filtering, (4) Summarization and (5) Design andPresentation. We outline these features in this section.5.1Generate Lesson PlanThe first component of the platform gives teachers the ability togenerate lesson plans by specifying a minimal set of keywords thatdescribe the contents that she wishes to cover in the respectivelesson.The teacher is presented with a simple HTML form which servesas the planning page for generating the lesson plan and consists ofdescriptive metadata, viz. Subject Name, Course title, Lesson titleand a few keywords describing the lesson outline. The teacher fillsin these details which initiates the lesson plan generation process;see Figure 1 for an example.Figure 1: Generate Lesson Plan InterfaceLimitations of CollectiveTeach 1.0Despite overall enthusiasm, teachers did express some skepticismthat inquiry-based approaches could take root in the current educational climate. Participant critiques of the platform from thesurveys and focus groups were collected and organized into threethemes: technical, structural, and cultural. Participants expressedthe following concerns about the CollectiveTeach 1.0 platform: Not being able to go back to previous stages and edit thelesson plan, especially once the results were returned as participants had new ideas for their plans but were not able toadd. (technical) Results from search not always useful, especially for theGhanaian context. Some resources are quite foreign to ourstudents. No local context. (cultural)5.1.1 Query Formulation. The addition of this feature to CollectiveTeach 2.0 increases the likelihood of improved search results.Based on the teacher-provided keywords, the platform formulatesspecific kinds of queries that aim to generate desirable and goodquality results. Specifically the teacher is not required to comeup with the most effective terms for improving the search results,which can be hard in practice [26]. The query formulation addsthree basic forms of structure to the search queries: (a) learning aset of domains relevant to a particular subject and constraining thesearch to pages within these domains; (b) targeting specific typesof files that provide highly relevant content for given subjects; (c)adding specific key words that improve relevance and quality ofsearch results.

Conference’17, July 2017, Washington, DC, USAAs an illustrative example, to fetch content from trusted sourceson the Web, we used search query formulation techniques such asappending “” and “” to the search queries,to help navigate the large space of educational content. Further,we also added targeted results for filetypes like PDFs and PPTs(Powerpoint) by appending the query with “filetype:pdf” and “filetype:ppt” respectively. We targeted the results to Wikipedia pagesfor the early stages in the 5E model, since they usually containcomprehensive description and explanation of a particular concept(and also its related concepts). Further, utilizing the knowledge thatprofessors around the world publicly post their course/lecture notesand conceptual explanations, we append terms such as ‘concepts’,‘notes’, etc. to further improve the results. Such queries tend toreturn both conceptual as well as application-based results. Similarly, for the Evaluate phase the query is modified with terms like‘homeworks’, ‘exams’, ‘midterm’, etc. which could return problemsets and exam papers from online educational websites. In general,the query formulation logic can be pre-encoded as a function ofthe subject using a small set of rules written by a user.5.1.2 Automated Filtering. Another addition to the CollectiveTeach2.0 platform is an automated filtering step that aims to discard irrelevant web links by applying empirically learnt heuristics. Forinstance, our lesson plan generator ignores links that only containthe course syllabus/outlines or course catalogs/schedules, as thementioned resources do not provide any “instructional content”.Other filtering techniques based on the content of the webpagescan be naturally incorporated.5.1.3 Presentation of Generated Results. To prevent user fatigueand/or overload, we aggregate the 5E model presented earlier intoonly two stages, which we refer to as Engage and Evaluate. This isrepresentative of college (or university) education where typicallyfor each course, there is a lecture each week that introduces andexplains some concepts, followed by some form of evaluation likeproblem sets, homeworks, assignments, etc. We also display the toplevel domain for each search result (like or well as the filetype (HTML, PDF, PPT etc.) to further enhancethe presentation. See Figure 2 for an example.Figure 2: Lesson Plan for Dynamic ProgrammingSave lesson plan button to add the lesson plan to her profile whereshe can view all the lesson plans that she has created.5.2Upload Lesson PlanCollectiveTeach 2.0 allows teachers to create their own lesson plansby uploading content in the form of documents, images/videos orother multimedia content. This helps in alleviating some of theconcerns raised by teachers in the user study earlier about webresources not providing the right local or cultural context. Further,teachers can also augment lesson plans generated using searchresults with their own (uploaded) content to obtain a lesson planthat most suits their need.5.3Search Lesson PlanA teacher can also search for existing lesson plans in the systemmade by other teachers and potentially incorporate relevant resources into her own lesson plans; see Figures 3 and 4. Note that5.1.4 Summarization of Generated Results. This feature aims atdisplaying a short summary of the webpage presented as part ofthe search result to help the teacher quickly gauge its relevancefor incorporating it as part of her lesson plan, as opposed to goingthrough the entire content of the webpage. Specifically, we usethe Summarization Search paradigm introduced by Chakraborty etal. [12] that performs a detailed text analysis of any search resultpage to prepare a condensed summary. The summarization involvesidentifying portions of the webpages that have high relevance (orsimilarity) with respect to the terms in the search query.5.1.5 Reordering and editing. The previous version of the platform did not allow teachers to go back and edit their lesson plansor reorder content within an existing lesson plan to match theirpresentation style. CollectiveTeach 2.0 provides both of these functionalities. Refer to Figure 2 which shows buttons at the extremeright of each search result that allow teachers to edit the generatedlesson plans. After this selection process, the teacher can press theFigure 3: Search Lesson Plan Interfacethis not only includes URLs but also documents and other educational material uploaded by teachers which might appear usefulfor improving the quality of the lesson plan. This feature of CollectiveTeach 2.0 allows teachers to share and re-use educational

Conference’17, July 2017, Washington, DC, USAA. Venkataraman et al.(2) Indexing and Search: Search works in conjunction withquery formulation and automated filtering described previously in Section 5.1. This extended version of the platformuses Elasticsearch1 to index, rank and search over the offline educational resources.(3) Fallback to online capability: In scenarios where theuser is not satisfied with the offline content returned byElastisearch, we can fetch additional content from the Web(assuming connectivity) as outlined above.content thereby promoting collaboration and resulting in a “collective teaching” environment.5.4.3 Multilingual Capability. Currently, the platform can support Spanish content (with English being the default). However, wecan easily extend the Generate Lesson Plan phase to fetch contentin a specific language (by modifying the parameters to the searchengine API). Similarly, the crawler will now need to index domainsbelonging to the provided language; we are currently working onthis.6Figure 4: Example Lesson plans returned via Search5.4ExtensionsIn addition to the

philosophy and psychology of Johann Herbart [23], which has a long history in educational theory grounded in ideas of Piaget and Dewey [29]. „e 5E model for preparing a lesson plan comprises of 5 stages: Engage, Explore, Explain, Elaborate and Evaluate. Col-lective